Cook>Book

NOTICE

This document contains confidential and proprietary information of Ab Initio Software Corporation.
Use and disclosure are restricted by license and/or non-disclosure agreements. You may not access,
read, and/or copy this document unless you (directly or through your employer) are obligated to
Ab Initio to maintain its confidentiality and to use it only as authorized by Ab Initio. You may not copy
the printed version of this document, or transmit this document to any recipient unless the recipient
is obligated to Ab Initio to maintain its confidentiality and to use it only as authorized by Ab Initio.

September 19, 2006 > Part Number AB0595

Ab Initio Software Corporation
201 Spring Street > Lexington MA 02421 > Voice 781.301.2000 > Fax 781.301.2001 > [email protected]

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 INTELLECTUAL PROPERTY RIGHTS & WARRANTY DISCLAIMER

CONFIDENTIAL & PROPRIETARY

This document is confidential and a trade secret of Ab Initio Software Corporation. This document is furnished under a
license and may be used only in accordance with the terms of that license and with the inclusion of the copyright notice set
forth below.

COPYRIGHTS
Copyright 1997-2006 Ab Initio Software Corporation. All rights reserved.
Reproduction, adaptation, or translation without prior written permission is prohibited, except as allowed under copyright
law or license from Ab Initio Software Corporation.

TRADEMARKS
The following are worldwide trademarks or service marks of Ab Initio Software Corporation (those marked are registered
in the U.S. Trademark Office, and may be registered in other countries):

> Continuous Flows Enterprise MetaEnvironment Plan>It

Ab Initio Continuous>Flows Everything 2 Everything Re>Posit
Ab Initio I>O Cooperating Enterprise Everything to Everything Re>Source
Abinitio.com Cooperating System From The Beginning Server++
Applications drawn to scale Cooperating GDE Server+Server
Co>Operating Enterprise Data>Profiler I>O Shop for Data
Co>Operating System Dynamic Data Mart If You Can Think It, You Can Do It The Company Operating System
Co>Operating EME init.com Think, Draw, Run, Scale, Succeed
Co>Operation EME Portal INIT
Co>Operative Enterprise Meta>Environment Meta Operating System
Co>OpSys Enterprise Metadata Environment Meta OS

Certain product, service, or company designations for companies other than Ab Initio Software Corporation are mentioned in
this document for identification purposes only. Such designations are often claimed as trademarks or service marks. In
instances where Ab Initio Software Corporation is aware of a claim, the designation appears in initial capital or all capital
letters. However, readers should contact the appropriate companies for more complete information regarding such
designations and their registration status.

ii AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

RESTRICTED RIGHTS LEGEND
If any Ab Initio software or documentation is acquired by or on behalf of the United States of America, its agencies and/or
instrumentalities (the Government), the Government agrees that such software or documentation is provided with
Restricted Rights, and is commercial computer software or commercial computer software documentation. Use,
duplication, or disclosure by the Government is subject to restrictions as set forth in the Rights in Technical Data and
Computer Software provisions at DFARS 252.227-7013(c)(1)(ii) or the Commercial Computer Software Restricted Rights
provisions at 48 CFR 52.227-19, as applicable. Manufacturer is Ab Initio Software Corporation, 201 Spring Street, Lexington,
MA 02421.

WARRANTY DISCLAIMER
THIS DOCUMENT IS SUBJECT TO CHANGE WITHOUT NOTICE. AB INITIO SOFTWARE CORPORATION MAKES NO WARRANTY OF ANY KIND WITH
REGARD TO THIS MATERIAL, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
PARTICULAR PURPOSE. AB INITIO SOFTWARE CORPORATION SHALL NOT BE LIABLE FOR ERRORS CONTAINED HEREIN OR FOR INCIDENTAL OR
CONSEQUENTIAL DAMAGE IN CONNECTION WITH THE FURNISHING, PERFORMANCE, OR USE OF THIS MATERIAL.

AB INITIO CONFIDENTIAL AND PROPRIETARY iii

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Contents
Getting started .............................................................................................. 1
Audience and prerequisites ...............................................................................................................2
Setting up..........................................................................................................................................2

Parsing names............................................................................................... 5
Graph for parsing names ...................................................................................................................6
Creating the graph .............................................................................................................................6

Generating large amounts of data ................................................................ 13

Simple data multiplier graph............................................................................................................14
Parameterized data multiplier graph................................................................................................15
Reusable, parameterized data multiplier subgraph ..........................................................................16
Other ways to multiply data .............................................................................................................17
Departitioners .........................................................................................................................18
Self-join with FILTER BY EXPRESSION to control output size ....................................................19

Making unique keys ..................................................................................... 21

Graph for making unique keys .........................................................................................................21
Globally unique ID ...................................................................................................................24
Unique ID within a single-field key group.................................................................................26
Unique ID within a two-field key group ....................................................................................28

AB INITIO CONFIDENTIAL AND PROPRIETARY Contents v

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Using a SCAN to generate unique keys ............................................................................................30
Using the SCANs key parameter to specify the key field(s) in the output.........................................31

Filling gaps................................................................................................... 33
Graph for filling gaps .......................................................................................................................34
Configuring the SCAN ......................................................................................................................36
The SCANs output .........................................................................................................................40
Writing the NORMALIZE transform ...................................................................................................41
Specifying the output record format.................................................................................................42
Running the graph and viewing the results ......................................................................................43

Measuring performance................................................................................ 45
Simple benchmarking graph ............................................................................................................45
Checking the accuracy of the outputs ..............................................................................................49
Other suggestions for measuring performance.................................................................................51
Startup time............................................................................................................................52
Elapsed time...........................................................................................................................53
Low-level operations...............................................................................................................54

Global sort.................................................................................................... 57
Simple global sort............................................................................................................................57
Sampling the data...................................................................................................................59
Calculating and broadcasting the splitters ..............................................................................59
Partitioning by range and sorting the partitions .......................................................................60
Checking the results ...............................................................................................................61
Producing serial output ...................................................................................................................62

Finding differences....................................................................................... 65
Creating a differencing graph...........................................................................................................68
Variations on simple differencing.....................................................................................................74

vi Contents AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 More complex differencing: Dealing with duplicate key values .........................................................74
Advanced differencing: Equality checks ...........................................................................................75
Safe equality checks ...............................................................................................................75
Fuzzy equality checks..............................................................................................................78

Automated generation of record formats and transforms.............................. 79

Generic differencing graph...............................................................................................................80
Metadata generation graph..............................................................................................................84
DML metaprogramming ...................................................................................................................85
Using DML to generate record formats ....................................................................................85
Using DML to generate transforms ..........................................................................................87

Quantiling .................................................................................................... 89
Simple serial deciling.......................................................................................................................90
Fixing decile boundaries ..................................................................................................................92
Parallel deciling...............................................................................................................................95
Getting the total record counts and starting record counts for each partition ..........................97
Calculating the splitters and fixing the boundaries across partitions .....................................101
Preparing the decile splitters for the final JOIN ......................................................................102
Assigning the deciles and outputting the data .......................................................................102
Parameterized parallel quantiling ..................................................................................................103
Creating a quantiling subgraph and inserting it into a test graph ...........................................104
Adding and exporting parameters ..........................................................................................106
Replacing hardcoded values with parameters .......................................................................107
Testing your quantiling subgraph...........................................................................................109

Auditing ..................................................................................................... 111

Sample application to audit ...........................................................................................................111
Simple auditing: Counting records .................................................................................................113

AB INITIO CONFIDENTIAL AND PROPRIETARY Contents vii

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Writing tracking information to a summary file and finding the flows needed for the audit.....113
Using a ROLLUP to check the record counts..........................................................................115
Limitations of using a summary file and a ROLLUP ...............................................................116
More complex auditing: Using log files...........................................................................................119
Collecting log data about the inputs ......................................................................................121
Rolling up the graphs output ................................................................................................124
Reading back and rolling up the values from the log files ......................................................125
Comparing the input and output values.................................................................................127
Testing the audit graph .........................................................................................................129

Processing sets of files............................................................................... 131

Ad hoc multifiles ...........................................................................................................................132
Listing files explicitly.............................................................................................................134
Using wildcards to list files ...................................................................................................135
Using a command to list files................................................................................................137
Using a parameter to list files ...............................................................................................137
Ad hoc output multifiles.................................................................................................................138
READ MULTIPLE FILES and WRITE MULTIPLE FILES components...................................................139
Reading multiple files ...........................................................................................................140
Reading and writing multiple files .........................................................................................144

Header and trailer record processing .......................................................... 151

Input data formats ........................................................................................................................152
Flat records ..........................................................................................................................153
Subrecords ...........................................................................................................................155
Vectors .................................................................................................................................158
Processing the data.......................................................................................................................160
Generating separate summaries of header and body subrecords...........................................160
Distributing header information to body records ....................................................................165

viii Contents AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Normalizing vector data to header and body records .............................................................167
Validating the input data................................................................................................................169
Validating flat input data .......................................................................................................169
Validating vector input data...................................................................................................174
Advanced topic: Building flat data into header/body/trailer data subrecord format.........................176
Rolling up to compute header and trailer data .......................................................................178
Reformatting to generic records.............................................................................................179
Merging to header, body, trailer order....................................................................................183
Reformatting to subrecord format..........................................................................................184

Geographic calculations ............................................................................. 185

Calculating distances between locations........................................................................................186
Haversine formula and great circle distance..........................................................................188
Simplifying the code and controlling the output by using the right data type ..........................190
Simple calculation of the nearest branch .......................................................................................193
Efficient calculation of the nearest branch .....................................................................................197
Building a lookup file of map grids.........................................................................................198
What the transform does.......................................................................................................199
Using the lookup file to find the nearest branch.....................................................................203

Cook>Book graphs listed by component

and feature ................................................................................................ 213
Graphs listed by component ..........................................................................................................213
Graphs listed by feature.................................................................................................................218

Index...................................................................................................................................................221

AB INITIO CONFIDENTIAL AND PROPRIETARY Contents ix

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Getting started
The Cook>Book presents a series of graphs that illustrate how you can use Ab Initio software to
solve a wide variety of problems. The solutions range from the very simple, where the emphasis
is on the mechanics of using the GDE (Graphical Development Environment), to the fairly
complex, where the algorithms are described in depth.
You can use the Cook>Book in two ways:
z As a source of recipes that you can either implement as is, modify slightly, or
re-create to solve a specific problem
z As a toolbox of running examples illustrating numerous features, tips, and techniques
that you can apply when developing your own Ab Initio graphs
Each chapter examines a different problem, and many of the chapters describe a series of graphs
that illustrate different approaches to solving it. The chapters are arranged roughly in order of
increasing complexity: the earlier ones provide detailed, step-by-step instructions (recipes) for
re-creating the graphs presented. Later chapters build on those, either by fully incorporating a
solution for a subproblem or by reusing a technique previously shown. The final chapters assume
that you have mastered the mechanics, and present guided tours of complete graphs, focusing
on highlights directly relevant to solving the problem at hand.

AB INITIO CONFIDENTIAL AND PROPRIETARY Getting started 1

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Audience and prerequisites
The Cook>Book is for those who have completed The Tutorial, the Introductory Course, and
Work>Book 1, and are eager to broaden and deepen their Ab Initio software development skills.

Listings of Cook>Book graphs by component and feature

For examples of specific components or features used in running graphs in the Cook>Book, see
the following sections:
z Graphs listed by component (page 213)
z Graphs listed by feature (page 218)

Setting up
To set up your Cook>Book environment:
1. Open a new instance of the GDE (to ensure that you will be running the setup script in the
host directory specified in the Run Settings dialog).
2. From the GDE menu bar, choose Run > Execute Command (or press F8).
3. In the Execute Command dialog, type the following and click OK:
$AB_HOME/examples/setup/set-up-cookbook.ksh

2 Getting started AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The setup script creates (in the current working directory) a directory named cookbook that
contains as subdirectories the various sandboxes referenced in this book. The script also builds
a multifile system (in AI_MFS) and data directories (in AI_SERIAL) for each sandbox, and runs
graphs to set up datasets.
The script can take anywhere from a few seconds to a few minutes, and displays Done when
finished.

AB INITIO CONFIDENTIAL AND PROPRIETARY Getting started 3

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Parsing names
String manipulation is one of the most common kinds of data transformation and is part of
almost every Ab Initio graph. The recipes in this chapter describe how to create a simple graph
that uses DML string functions and local variables to parse a name into separate strings for first
name, middle name or initial, and last name. For example, it reformats this name field:

to this:

AB INITIO CONFIDENTIAL AND PROPRIETARY Parsing names 5

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Graph for parsing names
The graph name-parsing.mp (in the appetizers sandbox) uses a REFORMAT component to parse
the input file:

The input file stores the first name, middle initial, and last name in the name field of the
following record format (in people.dml):
record
decimal(',') id;
string(',') name;
decimal('\n') age;
end

Creating the graph

To create a graph like name-parsing.mp:
1. Add an INPUT FILE with a record format such as the one shown above, which stores an entire
name in a single field, and an input dataset such as people.dat.

6 Parsing names AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

2. Add an OUTPUT FILE, specify a name for it, and embed the following record format:
record
decimal(',') id;
string(',') last_name;
string(',') first_name;
string('\n') middle;
decimal('\n') age;
end
3. Add a REFORMAT component between the input and output files, and connect them to it.
4. In the REFORMAT component, use the Transform Editors Grid View to create the local
variables listed in the following table, as described in the instructions after the table:

VARIABLE NAME TYPE LENGTH EXPLANATION AND VALUE

tname string "" Holds in.name with leading and trailing blank characters
removed:
string_lrtrim(in.name)

end_of_first integer 4 Represents the location of the first blank character in tname:
string_index(tname, ' ')

beginning_ of_last integer 4 Represents the location of the last blank character in tname:
string_rindex(tname, ' ')

a. Hold down the Shift key and double-click the REFORMAT component to display the
Transform Editor.

NOTE: If the Transform Editor opens in Text View, choose View > Grid View.
b. Switch to the Variables tab and choose Edit > Local Variable to display the Variables
Editor.
c. On the Fields tab, enter the name, type, and length attributes shown above, and save
the variable.

AB INITIO CONFIDENTIAL AND PROPRIETARY Parsing names 7

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 d. On the Variables tab of the Transform Editor, type the following into the first blank
line, or drag the name field from the Inputs pane into the first blank line and type the
rest of the code around it:
string_lrtrim(in.name)
e. Connect the rule port to the variable tname in the Outputs pane.
If you view the transform in Text View, it should now contain the following line of code:
let string("") tname =string_lrtrim(in.name);
f. Create the other two variables in the same way, each time using the next blank line of
the Variables tab.
As you do this, notice that the Variables section of the Inputs pane incrementally
shows the previously created variable(s), indicating that they are available for use in
creating additional variables.
If you view the transform in Text View after creating the three variables, you should see
the following code:
out::reformat(in) =
begin
let string("") tname =string_lrtrim(in.name) ;
let integer(4) end_of_first =string_index(tname, ' ');
let integer(4) beginning_of_last =string_rindex(tname, ' ');
end;

8 Parsing names AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

5. Create three business rules to construct the output fields (listed in the following table) for the
three parts of the name. You can do this in the Transform Editors Grid View, as described in
the instructions after the table.

OUTPUT FIELD DESCRIPTION BUSINESS RULE

out.last_name Extends from the character after the last string_substring(tname,

blank to the end of the string beginning_of_last + 1,
length_of(tname))

out.first_name Extends from the start of name to the string_substring(tname, 1,

last character before the first blank end_of_first - 1)

out.middle Extends from the first character after the string_substring(tname,

first blank to the character before the end_of_first + 1,
last blank beginning_of_last -
end_of_first - 1)

a. Position the cursor in the first blank line on the Business Rules tab.
b. Choose Edit > Rule to display the Expression Editor.
c. In the Functions pane, expand the String Functions item and double-click
string_substring(str, start, length) to display an empty function definition
(string_substring(?, ?, ?)) in the bottom pane.
d. Replace the question marks with the following parameters, either by typing them
directly or by dragging the variable names from the Fields pane and typing as needed:
string_substring(tname, beginning_of_last + 1,
length_of(tname))
e. Click OK to return to the Transform Editor.
f. Drag the rule port to the last_name field in the Outputs pane.

AB INITIO CONFIDENTIAL AND PROPRIETARY Parsing names 9

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 g. Create two more business rules in the same way to construct the output fields for
first_name and middle.
If you view the transform in Text View after creating the three variables, you should see
the following additional code:
out.last_name :: string_substring(tname,
beginning_of_last + 1, length_of(tname));
out.first_name :: string_substring(tname, 1,
end_of_first - 1);
out.middle :: string_substring(tname, end_of_first + 1,
beginning_of_last - end_of_first - 1);

NOTE: In the statement that computes last_name, the call to string_substring uses
the length of the entire string length_of(tname) for the length of the substring.
This has the effect of simply taking all the characters starting at beginning_of_last +
1. While the length calculation could have been refined (to length_of(tname)
beginning_of_last), the result would be the same, as string_substring simply returns
the available characters.

10 Parsing names AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

6. Assign the id field and the age field from the inputs to the outputs: in the Transform Editors
Grid View, drag each field from the Inputs pane over the first empty line at the bottom of the
Business Rules tab and onto the field with the corresponding name in the Outputs pane.

7. Run the graph and view the results.

AB INITIO CONFIDENTIAL AND PROPRIETARY Parsing names 11

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Generating large amounts of data GENERATE RECORDS
COMPONENT
The GENERATE RECORDS component
To test performance, you should run every graph on a volume of records that is comparable to is good way to generate random
what you expect in production. But when developing and testing a graph, you may have only a records, but if you need very large
small sample of real data, or none at all. The recipes in this chapter describe how to create volumes of data, you should use it
graphs that generate large amounts of test data efficiently. with one of the techniques
described here to get best
The multiplying-data sandbox contains graphs that use a NORMALIZE component to generate
performance.
large amounts of data by doing the following:
z Multiplying records by a fixed number (page 14)
z Multiplying records by a parameterized number (page 15)
z Multiplying records by a parameterized number in a reusable subgraph (page 16)
Two additional graphs demonstrate how to generate data by using other components:
z Departition components: CONCATENATE, GATHER, and INTERLEAVE (page 18)
z Three-way self-join on an empty key with a FILTER BY EXPRESSION component (page 19)

AB INITIO CONFIDENTIAL AND PROPRIETARY Generating large amounts of data 13

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Simple data multiplier graph
The graph multiplying-data-by-fixed-number.mp multiplies the four records in the input file
1,000,000 times, producing 4,000,000 records:

To create a graph like multiplying-data-by-fixed-number.mp:

1. Attach a NORMALIZE component to your input file.
2. Hold down the Shift key and double-click the NORMALIZE component to generate a default
transform.
3. Delete the temporary_type variable and all default functions except length and normalize.
4. Code the length function (which determines the number by which to multiply the input
records) as follows, assigning the multiplier to the out record:
out :: length(in) =
begin
out :: 1000000;
end;
5. Code the normalize function to simply assign the in records to each out record:
out :: normalize(in, index) =
begin
out :: in;
end;

14 Generating large amounts of data AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

6. Send the generated data to the component you need to test.

NOTE: The REFORMAT component in the graph is not explicitly described here, as it is merely
a placeholder representing any component or set of components that need to be tested
with a large volume of data.

Parameterized data multiplier graph

The graph multiplying-data-with-parameter.mp multiplies the records in the input file by a
number specified in a graph parameter. The parameter is multiplier and its value is 1000000.

To create a graph like multiplying-data-with-parameter.mp:

1. Create a simple data multiplier graph as described on page 14.
2. Add a graph parameter (such as multiplier).
3. Set the interpretation of the NORMALIZE components transform parameter to
${} substitution.

AB INITIO CONFIDENTIAL AND PROPRIETARY Generating large amounts of data 15

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 4. In the length function of the NORMALIZE components transform, replace the hardcoded value
with the parameter as follows:
out :: length(in) =
begin
out :: ${multiplier};
end;

Reusable, parameterized data multiplier subgraph

The graph multiplying-data-with-parameterized-subgraph.mp contains a reusable,
parameterized subgraph. (The parameter is multiplier and its value is 1000000.)

To create a graph that contains a reusable subgraph (like multiplying-data-with-

parameterized-subgraph.mp):
1. Create a parameterized data multiplier graph as described on page 15.
2. Create a subgraph from the NORMALIZE component by right-clicking the component and
choosing Subgraph > Create From Selection.

16 Generating large amounts of data AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

3. Delete the multiplier parameter from the graph level as follows:
a. Choose Edit > Parameters.
b. In the Edit Parameters dialog, select the multiplier parameter in the Name field, and
choose Edit > Delete.
4. Create a multiplier parameter at the subgraph level:
a. Select the subgraph and choose Edit > Parameters.
b. In the Name field of the Edit Parameters dialog, type multiplier.
c. In the Value field of the Edit Parameters dialog, type the value you want.
d. Save the changes to the parameters.

Other ways to multiply data

A NORMALIZE-based data multiplier is used in the preceding recipes because it is very efficient
and completely generic. However, you can also create large amounts of data quickly by using a
departitioner or a self-join on an empty key, as described in the following sections:
z Departitioners (next)
z Self-join with FILTER BY EXPRESSION to control output size (page 19)

AB INITIO CONFIDENTIAL AND PROPRIETARY Generating large amounts of data 17

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Departitioners
You can multiply data by using a departition component, as shown in the graph multiplying-
data-with-departitioners.mp:

To generate M copies of the input data:

z Connect an input file to one of the following components with M flows:
CONCATENATE To get M copies of the whole file, one after another
GATHER To get M copies of each record, in some arbitrary order
INTERLEAVE To get M copies of each record in succession

18 Generating large amounts of data AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Self-join with FILTER BY EXPRESSION to control output size
With a two-way self-join on an empty key, you can generate N*N output records from N input
records. With a three-way self-join, you can generate N*N*N output records. You can further
control the output size by sending one of the inputs through a FILTER BY EXPRESSION before it
enters the JOIN. As you can see in the graph multiplying-data-by-3-way-self-join-with-FBE.mp,
several techniques can be applied in tandem to multiply data:

To create a graph like multiplying-data-by-3-way-self-join-with-FBE.mp:

1. Connect multiple flows of your input file to a GATHER component.
2. Replicate the output of the GATHER.
3. Send one flow from the REPLICATE to a FILTER BY EXPRESSION component, and configure the
FILTER BY EXPRESSION as follows:

a. Hold down the Shift key and double-click the FILTER BY EXPRESSION to display the
Expression Editor.

AB INITIO CONFIDENTIAL AND PROPRIETARY Generating large amounts of data 19

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

USING THE random FUNCTION b. In the Functions pane, double-click Miscellaneous Functions to expand the list.
IN A FILTER BY EXPRESSION TO c. Double-click random(int) returns long to display random(?) in the bottom pane of the
CONTROL OUTPUT SIZE Expression Editor.
For each incoming record, the
d. To control the percentage of the records entering the FILTER BY EXPRESSION that will be
random function generates a
random integer in the range from 0 passed to its out port, edit the expression to something like this (see the sidebar for
to the value passed to it. By details about the random function):
passing 100 to the random function random(100) < N
and then testing whether the where N is the percentage of records (in this graph, N is 50).
generated number is less than a
4. Connect the out port of the FILTER BY EXPRESSION and the remaining flows from the REPLICATE
specified value (in an expression
such as random(100) < N), you to a JOIN, and configure the JOIN as follows:
cause the FILTER BY EXPRESSION to a. Set the key to the empty key: { }
send approximately N percent of
b. Set the transform parameter in the Transform Editors Grid View by dragging in0 from
the incoming records to its out port.
the Inputs pane over the first empty line in the Business Rules tab and onto out0 in the
Outputs pane.
If you view the transform in Text Mode after doing this for a three-way join, you should
see the following code:
out::join(in0, in1, in2) =
begin
out :: in0;
end;
5. Connect the JOIN to an OUTPUT FILE.

20 Generating large amounts of data AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Making unique keys
In some situations you may need to assign unique keys to data that is not uniquely keyed to
begin with. The recipes in this chapter describe how to use the SCAN component to generate
unique keys globally or within an existing key grouping.

Graph for making unique keys

The graph making-unique-keys.mp (in the appetizers sandbox) contains three SCAN
components that use the same transform code and the same output record format. But because
each SCAN uses a different key, the scope of uniqueness is different for each output file.

AB INITIO CONFIDENTIAL AND PROPRIETARY Making unique keys 21

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The input record format for the graph is specified by events.dml:
record
date("YYYY-MM-DD")(",") dt;
string(",") kind;
decimal("\n") amount;
end

The input file contains the following data, which is sorted on the date field (dt):

22 Making unique keys AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

The output record format contains one additional field, id, and is specified by events-keyed.dml:
record
decimal(",") id;
date("YYYY-MM-DD")(",") dt;
string(",") kind;
decimal("\n") amount;
end

AB INITIO CONFIDENTIAL AND PROPRIETARY Making unique keys 23

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Each SCAN generates a count value and assigns it to the id field in the output file, but each does
this in a different way, as described in the following sections.

Globally unique ID
The key parameter of the Scan: Assign id Globally component is set to the empty key: { },
causing the entire input file to be treated as a single key group. The SCAN simply counts the
records and assigns the current count to the id field for each record, so that every record is
uniquely identified across the entire output file. The output data looks like this:

24 Making unique keys AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

AB INITIO CONFIDENTIAL AND PROPRIETARY Making unique keys 25

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Unique ID within a single-field key group
The key parameter of the Scan: Assign id Within {dt} component is set to the date field, {dt},
causing each record with the same dt value to be treated as part of the same key group. The
SCAN counts the records in each key group and assigns the current count within that key group
to the id field for each record. In the output, the combination of the dt field and the id field
uniquely identifies each record. The output data looks like this:

26 Making unique keys AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

AB INITIO CONFIDENTIAL AND PROPRIETARY Making unique keys 27

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Unique ID within a two-field key group
The Scan: Assign id Within {dt; kind} component has a two-part key parameter {dt; kind}, so it
treats each record with the same pair of dt and kind values as part of the same key group. Again,
it counts the records in each key group and assigns the current count within that key group to
the id field for each record. In the output, the combination of the three fields dt, kind, and id
uniquely identifies each record.

NOTE: Like the other two SCAN components in the graph, this is a sorted SCAN (its sorted_input
parameter is set to Input must be sorted or grouped). The other two SCANs work without
upstream sorting, because the input data was already sorted on the dt field, and key was set
to either the empty key: { } or the dt field. But because the third SCAN has a two-field key
parameter, {dt; kind}, the transform will fail if the input data is not sorted on both of those
fields. Therefore, this SCAN must be preceded by a SORT WITHIN GROUPS component that uses
dt as its major_key parameter and kind as its minor_key parameter.
The output data looks like this:

28 Making unique keys AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

AB INITIO CONFIDENTIAL AND PROPRIETARY Making unique keys 29

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Using a SCAN to generate unique keys
To create a graph with a SCAN that generates unique keys for the records in a dataset, like the
SCAN components in making-unique-keys.mp:

NOTE: The scope of uniqueness for the count value generated by the SCAN and assigned to an
output field is determined by the key parameter of the SCAN, as described in Using the
SCANs key parameter to specify the key field(s) in the output (page 31).
1. Attach a SCAN component to your input file.
2. Hold down the Shift key and double-click the SCAN component to display the Transform
Editor.

NOTE: If the Transform Editor opens in Grid View, choose View > Text.
3. Delete all default transform functions except initialize, scan, and finalize.
4. Define the type temporary_type to be a decimal("") with an appropriate name (such as
count) by editing the default definition to read as follows:
type temporary_type =
record
decimal("") count;
end;
5. Code the initialize function as follows to initialize count to zero:
temp::initialize(in) =
begin
temp.count :: 0;
end;
6. Code the scan function as follows to increment count by 1 for each record it processes:
temp::scan(temp, in) =
begin
temp.count :: temp.count + 1;
end;

30 Making unique keys AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

7. Code the finalize function as follows so that for each record, the SCAN will assign the value of
count to the appropriate output field (for example, id) and will assign all other input fields to
the output fields with the same names:
out::finalize(temp, in) =
begin
out.id :: temp.count;
out.* :: in.*;
end;

Using the SCANs key parameter to specify the key field(s) in

the output
If you use a SCAN transform like the one in making-unique-keys.mp to generate a unique
identifier, the key parameter of the SCAN component determines the scope within which the
identifier is unique, and thus also determines the key field or fields that uniquely identify each
record in the output.

AB INITIO CONFIDENTIAL AND PROPRIETARY Making unique keys 31

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 TO SPECIFY THIS SCOPE WHICH MEANS SET THE KEY PARAMETER TO THIS

Global Every record across the entire output dataset is The empty key: { }
uniquely identified by the output field to which
the SCANs count field is assigned. In the
output for the first SCAN in making-unique-
keys.mp, the unique key is the id field.

A specific key group Each record within a key group is uniquely The field or fields that form the
identified by the combination of the field(s) key of the key group
specified in the key parameter, and the output NOTE: If the input file is not
field to which the SCANs count field is already sorted on this field or
assigned: these fields, you can use a
z In the output for the second SCAN in SORT or SORT WITHIN
making-unique-keys.mp, the unique key GROUPS as necessary, or use
is the combination of the dt and id fields. an in-memory SCAN instead.
z In the output for the third SCAN in
making-unique-keys.mp, the unique key
is the combination of the dt, kind, and id
fields.

32 Making unique keys AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Filling gaps
Sometimes data is provided at a coarser granularity than is required for a given business task. For
example, your input data may contain account balances only for days with activity, but you may
need to process that data at the daily level. The input data is effectively missing records for
days on which there was no account activity.
The recipe in this chapter demonstrates how to generate daily balances for days or sequences of
days on which an account showed no activity and for which the input data therefore lacks
records.

AB INITIO CONFIDENTIAL AND PROPRIETARY Filling gaps 33

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Graph for filling gaps
The graph fill-gaps.mp (in the appetizers sandbox) does this by using the following components:
z SCAN Detects each gap in the input data and calculates the span of the gap (that is,
the number of days for which the graph needs to generate balances). The SCANs output
also preserves the last balance preceding each gap, so that it can be used in the new
records.
z NORMALIZE Uses the spans to generate the correct number of records to fill each
gap, and carries the balance from the last record preceding each gap into the generated
records.

34 Filling gaps AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

The input file for fill-gaps.mp contains the following data, which has gaps between several of
the dates:

AB INITIO CONFIDENTIAL AND PROPRIETARY Filling gaps 35

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Configuring the SCAN
This section describes how to configure a SCAN to detect each gap in the input records and
calculate the span of the gap.
To configure the SCAN:
1. Set the key to the empty key: { } so that all records belong to the same key group.
2. Calculate the length of the gaps as follows:
a. In the SCANs temporary_type, initialize this_date and prev_date to a special value,
such as the empty string:
type temporary_type =
record
date("YYYY.MM.DD") this_date = '';
date("YYYY.MM.DD") prev_date = '';
...

NOTE: Using a special value allows you to treat the first record differently (see the
finalize function in Step c).
b. In the scan function, write the current balance date to the this_date field of the current
temporary record (for use in processing the next record), and the this_date field of the
previous temporary record to the prev_date field of the current temporary record:
out :: scan(temp, in) =
begin
out.this_date :: in.bal_date;
out.prev_date :: temp.this_date;
...

36 Filling gaps AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 c. In the finalize function, output the in.bal_date and calculate the span by subtracting
the date of the previous record from the date of the current record (or returning 1 if the
previous record has an invalid date):
out :: finalize(temp, in) =
begin
out.* :: in.*;
...
out.span :1: if (not is_blank(temp.prev_date))
in.bal_date - temp.prev_date;
out.span :2: 1;
...
The following table shows what this code does with the first four records:

TEMPORARY VALUE OUTPUT

RECORD in.bal_date this_date prev_date SPAN

1 2006.01.03 2006.01.03 "" 1

2 2006.01.04 2006.01.04 2006.01.03 1

3 2006.01.08 2006.01.08 2006.01.04 4

4 2006.01.12 2006.01.12 2006.01.08 4

3. Calculate the values needed to fill the gaps as follows:

a. In the SCANs temporary_type, initialize this_amount and prev_amount to zero:
type temporary_type =
record
...
decimal(10.2) this_amount = 0;
decimal(10.2) prev_amount = 0;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Filling gaps 37

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. In the scan function, write the current balance amount to the this_amount field of the
current temporary record (for use in processing the next record), and the this_amount
field of the previous temporary record to the prev_amount field of the current
temporary record:
out :: scan(temp, in) =
begin
...
out.this_amount :: in.amount;
out.prev_amount :: temp.this_amount;
end;
c. In the finalize function, output in.amount and prev_amount:
out :: finalize(temp, in) =
begin
out.* :: in.*;
out.prev_amount :: temp.prev_amount;
...
The following table shows what this code does with the first four records:

TEMPORARY VALUE OUTPUT

RECORD in.amount this_amount prev_amount amount prev_amount

1 345.92 345.92 0 345.92 0

2 1234.23 1234.23 345.92 1234.23 345.92

3 234.12 234.12 1234.23 234.12 1234.23

4 12334.22 12334.22 234.12 12334.22 234.12

38 Filling gaps AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Following is the complete transform code for the SCAN:
type temporary_type =
record
date("YYYY.MM.DD") this_date = '';
date("YYYY.MM.DD") prev_date = '';
decimal(10.2) this_amount = 0;
decimal(10.2) prev_amount = 0;
end;

out :: scan(temp, in) =

begin
out.this_date :: in.bal_date;
out.prev_date :: temp.this_date;
out.this_amount :: in.amount;
out.prev_amount :: temp.this_amount;
end;

out :: finalize(temp, in) =

begin
out.* :: in.*;
out.prev_amount :: temp.prev_amount;
out.span :1: if (not is_blank(temp.prev_date))
in.bal_date - temp.prev_date;
out.span :2: 1;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Filling gaps 39

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The SCANs output
If you want to view the output of the SCAN, choose Debugger > Enable Debugger, right-click the
out flow, and click Watcher on Flow. The output should look like this:

40 Filling gaps AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Writing the NORMALIZE transform
The NORMALIZE component reads the output of the SCAN and uses it to generate the missing
records. The NORMALIZE reads the span field to determine how many output records are needed
to fill each gap, generates the records, and writes the value from prev_amount to the amount
field as necessary to fill the gaps.
To generate the missing records with the appropriate values:
1. In the NORMALIZEs length function, use the span field from the SCAN component to specify
the number of output records needed to fill the gap between the current and the next input
record:
out :: length(in) =
begin
out :: in.span;
end;
2. In the NORMALIZEs normalize function, generate a record for each date in the gap as follows:
a. Calculate the date by subtracting the difference between the entire span and the index
into it from the current balance date, and adding 1:
out :: normalize(in, index) =
begin
out.bal_date :: in.bal_date - (in.span - index) + 1;
This returns subsequent dates in the range covering the gap.
b. Use the previous amount for records in the gap (index < in.span - 1), and the current
amount for the final record:
out.amount :1: if (index < in.span - 1) in.prev_amount;
out.amount :2: in.amount;
out.* :: in.*;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Filling gaps 41

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Specifying the output record format
To specify the output record format:
1. Attach an OUTPUT FILE component to the SCAN.
2. Specify the URL to a file containing the following record format, or embed it directly into the
OUTPUT FILE:
record
date("YYYY.MM.DD") bal_date;
decimal(10.2) amount;
string(1) newline = '\n';
end

42 Filling gaps AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Running the graph and viewing the results
When you run the graph and view the results, the first 20 records should look like this:

AB INITIO CONFIDENTIAL AND PROPRIETARY Filling gaps 43

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Measuring performance
When building graphs in Ab Initio software, you will often discover that you can accomplish the
same task in more than one way. In deciding on the best approach, you may need to consider
various factors, and although clarity and maintainability may be paramount in some cases,
performance will often be a concern. However, before you can make an informed decision about
structuring your graph to maximize performance, you must have a methodology for measuring
it.
The recipes in this chapter explain how to compare the performance of various approaches to a
task while minimizing or eliminating factors that could distort the comparison. It describes the
following:
z A simple benchmarking graph (next)
z A benchmarking graph that checks its results (page 49)
z Other suggestions for measuring performance (page 51)

Simple benchmarking graph

The graph benchmarking.mp (in the appetizers sandbox) measures the speed of three different
transform functions that test the validity of account numbers. It minimizes the effects of the
following factors:

AB INITIO CONFIDENTIAL AND PROPRIETARY Measuring performance 45

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 z Startup time Graphs can spend a significant amount of time just starting up. This
graph minimizes the effect of startup time by using the Multiply Data subgraph to
generate a large amount of data. For details on this subgraph, see Reusable,
parameterized data multiplier subgraph (page 16).
z Environmental factors Factors such as system load can vary between runs of a graph,
thereby affecting the accuracy and repeatability of performance measurements. This
graph eliminates such factors by running all three transform functions on replicated
data in the same graph.

46 Measuring performance AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

To create a graph like benchmarking.mp: TEXT TRACKING
1. If your input dataset does not contain enough records to test different approaches to the task, Text tracking provides CPU utilization
attach a Multiply Data subgraph to your input dataset, and set its multiplier parameter times to the precision of hundredths
appropriately. (and sometimes thousandths) of a
second, whereas the tracking
In benchmarking.mp, the test data consists of only four records, and multiplier is set to information in the GDE Tracking
500000. For details on creating a Multiply Data subgraph, see Reusable, parameterized data Details dialogs is often rounded to the
multiplier subgraph (page 16). nearest second, possibly obscuring
2. Attach a REPLICATE component so that you can replicate the data once for each approach you small differences in performance. For
this reason, text tracking is
want to test.
sometimes a better tool for examining
3. Attach the appropriate component for each test to a separate flow from the REPLICATE, and performance.
code the components appropriately. When using Version 2.14 (or higher)
In benchmarking.mp, three REFORMAT components each use a different transform to validate of the Co>Operating System and
the account numbers. For details on the transforms, see About the REFORMAT components Version 1.14 (or higher) of the GDE,
in benchmarking.mp (page 48). you can display both text tracking and
GDE tracking (in the Tracking Details
4. Send the resulting data to a TRASH component. dialog). With earlier versions, you
In this case simply discard the output. When you need to be sure that different approaches must choose between text tracking
lead to the same results, you should also check the accuracy of the outputs as described in and the Tracking Details dialog.
Checking the accuracy of the outputs (page 49). To turn on text tracking:
5. If appropriate, turn on Text Tracking (see the sidebar for details). 1. Choose Run > Settings.
6. Run the graph and view the tracking data for each component. 2. Click the Tracking tab.
3. Select the Text Tracking
The section Phase 0 ended shows that the Reformat: Test Using is_valid component uses less checkbox.
CPU time than the other two REFORMAT components: Text tracking appears in the Job
----------------------------------------------------------------- Output dialog as the graph runs.
Feb 13 17:17:24 Phase 0 ended (14 seconds)
CPU Time Status Skew Vertex
3.359 [ : 1] 0% Reformat_Test_Using_is_valid
5.453 [ : 1] 0% Reformat_Test_Using_re_index

AB INITIO CONFIDENTIAL AND PROPRIETARY Measuring performance 47

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 4.141 [ : 1] 0% Reformat_Test_Using_string_filter_out
1.078 [ : 1] 0% Replicate
1.859 [ : 1] 0% Subgraph_Multiply_Data.Normalize
0.313 [ : 1] 0% Trash_A
0.422 [ : 1] 0% Trash_B
0.297 [ : 1] 0% Trash_C
-----------------------------------------------------------------

About the REFORMAT components in benchmarking.mp

In benchmarking.mp, each of the three REFORMAT components uses a different transform
function to test the validity of the account_num field in the input dataset:
z re_index Uses regular expression matching to check whether the account number
consists only of digits:
out::reformat(in) =
begin
out.account_num :1: if (re_index(in.account_num,
"[^0-9]+") == 0)
in.account_num;
out.account_num :: 0;
out.* :: in.*;
end;
z is_valid Checks whether the account number is a valid decimal number:
out::reformat(in) =
begin
out.account_num :1: if (is_valid(in.account_num))
in.account_num;
out.account_num :: 0;
out.* :: in.*;
end;

48 Measuring performance AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 z string_filter_out Checks whether each character in the account_num field is a digit:
out::reformat(in) =
begin
out.account_num :1: if (length_of(string_filter_out(
in.account_num, "0123456789")) == 0)
in.account_num;
out.account_num :: 0;
out.* :: in.*;
end;

Checking the accuracy of the outputs

Obviously it does not make sense to compare the performance of transforms whose accuracy is
questionable. However, if you do an initial comparison and then change your benchmarking
graph to improve the performance of any of its transforms, you may also inadvertently change
its behavior in unintended ways. To recheck the accuracy of your transforms, create a graph
such as benchmarking-with-checking.mp:

AB INITIO CONFIDENTIAL AND PROPRIETARY Measuring performance 49

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 This graph uses a FUSE component as a quick and easy way to check specifically for differences
between the values of the account_num field in the outputs of the previous graphs two fastest
tests, Test Using is_valid and Test Using string_filter_out. To check for more general differences
(such as added, deleted, or changed records), see Finding differences (page 65) and
Automated generation of record formats and transforms (page 79).

50 Measuring performance AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

In Text View, the FUSE components transform looks like this: ABOUT RECORD ORDER
out::fuse(in0, in1) = REFORMAT components do not change
begin the order of records, so you can use a
out :: if (in0.account_num != in1.account_num) FUSE to compare the results of the
force_error( different transforms. If you are
string_concat("Mismatch in account_num (in0='", measuring the performance of
in0.account_num, "'; in1='", collections of components that may
in1.account_num, "')")) result in a different ordering of data, it
else
may be appropriate to sort before a
in0;
FUSE or use the more general data
end;
differencing technique described in
If you run this graph, you will see that Test Using is_valid accepts the account number 45.8, Finding differences (page 65) and
whereas Test Using string_filter_out catches it and sets the account_num field to zero: Automated generation of record
Error from Component 'Fuse', Partition 0 formats and transforms (page 79).
[U119]
Exceeded reject threshold after 1 rejected records.
While executing transform fuse:

Mismatch in account_num (in0=' 45.8'; in1=' 0') in assignment

for out.

Other suggestions for measuring performance

The graph benchmarking.mp focuses on CPU time, and it measures the performance of
transforms in individual components. You can use similar techniques to measure the
performance of collections of components in the same graph.
To measure the performance of different approaches when you cannot use the same graph:
1. Do at least three runs of each approach, perhaps alternating them.

AB INITIO CONFIDENTIAL AND PROPRIETARY Measuring performance 51

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 2. Take the minimum time for each approach. Use the minimum (instead of the average, for
example) because environmental factors may make a graph slower, but they will never
make it faster.

Startup time
The effect of startup time is proportionally larger when processing small data than when
processing large data, so you may need to take it into account when measuring or estimating
performance. Following is a description of how to do this for the sample record numbers and
times listed in the table below.

TIME (SEC)

RECORDS MEASURED (ACTUAL) SIMPLE PROJECTION PROJECTION WITH STARTUP

1 3.15 3.15

50,000 4.35 4.35 4.35

500,000 43.50 15.15

To project the time (incorporating startup time) for a run on big data:
1. Run the graph on very small data (one record) to get the startup time (here 3.15 sec).
2. Run the graph on small data (here 50,000 records take 4.35 sec).
3. Subtract the startup time from the total time for small data to get a scaling factor for use
with large data (here 4.35 - 3.15).
4. Multiply the ratio of big data to small data by the scaling factor, and then add back the
startup time for example:
(500,000/50,000) * (4.35 3.15) + 3.15 = 15.15

52 Measuring performance AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Taking startup time into account makes performance projections much more accurate. The
following chart shows that the difference between a projection that incorporates startup time
and one that does not can be huge.

Elapsed time
You can measure elapsed time in different ways. Since elapsed time is sensitive to load, you
should run on an unloaded system and/or do multiple runs and take the minimum, as described
earlier. You can get information on elapsed time in several ways:
z Elapsed time for an entire graph Displayed on the bottom right, in the GDE status
bar. If the status bar is not visible, choose View > Toolbars > Status Bar.

AB INITIO CONFIDENTIAL AND PROPRIETARY Measuring performance 53

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 z Elapsed time per phase Provided in these ways:
As part of text tracking, which outputs the performance data for the graph. (For
information on text tracking, see Text tracking on page 47.)
In the graphs summary file. For details on the summary file, see Writing tracking
information to a summary file and finding the flows needed for the audit
(page 113) and Ab Initio Help.

Low-level operations
To measure the cost of a very-low-level operation (processing you might do multiple times per
record or field) and factor out per-record processing and I/O overhead, use a for loop to execute
the operation N times per record. The graph benchmarking-low-level.mp uses this technique to
compare the performance of two different DML functions that extract strings with a specific
pattern.

54 Measuring performance AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

To create a graph like benchmarking-low-level.mp: TO ADD AND SET AN INPUT
1. Add an input parameter (for example, REPEAT_AMT; see the sidebar for details), and set its PARAMETER
value to the number of times the low-level operation must be performed to eliminate the 1. Choose Edit > Parameters.
effects of per-record processing and I/O overhead. In benchmarking-low-level.mp, the 2. In the Name field of the Edit
parameter is set to 500000. Parameters dialog, type a name
for the parameter (for example,
To determine a reasonable value, start with a low number and increase it until the weight of REPEAT_AMT).
the other factors is low enough that the processing time increases almost exactly in step with 3. In the Value field, type a value for
REPEAT_AMT. the parameter (for example,
500000).
2. Add an input file containing your test data.
3. Attach two REFORMAT components to the input file and code the transform in each to DO NOT USE REPLICATE
perform the desired operation the number of times specified in the input parameter. COMPONENTS
For example, the transforms in the two REFORMAT components of benchmarking-low- UNNECESSARILY
level.mp are coded as follows (the benchmarked code is underlined):
When you need to read data from a
file multiple times before
Reformat: Use re_get_match processing it in any way, simply
out::reformat(in) = attach multiple flows to the file, as
begin in benchmarking-low-level.mp. This
let string("") result = ""; allows downstream components to
let integer(4) i = 0; run as fast as they can, without
for (i, i < ${REPEAT_AMT}) waiting for others to receive their
result = re_get_match(in.name, "Bad.*"); copy of the data from the REPLICATE.
out.* :: in.*;
While the effect may be small in
out.name :1: result;
simple graphs such as these, it can
out.name :: "";
be significant in larger graphs, so
end;
avoiding REPLICATE in all such
Reformat: Use string_index and string_substring situations is a good practice.
out::reformat(in) =
begin
let string("") result = "";
let integer(4) i = 0;
for (i, i < ${REPEAT_AMT})

AB INITIO CONFIDENTIAL AND PROPRIETARY Measuring performance 55

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 if (string_index(in.name, "Bad") > 0)
result = string_substring(in.name,
string_index(in.name, "Bad"),
string_length(in.name));
out.* :: in.*;
out.name :1: result;
out.name :: "";
end;
4. Run the graph.
5. When the graph finishes, right-click anywhere on the graph and choose Tracking Detail.
The information in the Cpu column shows that the expression using string_index and
string_substring is more efficient for this particular operation.

56 Measuring performance AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Global sort
While it is quite common to maintain sorted data in multifile datasets, such data is usually
sorted only within each partition, not across partitions. In some situations, however, a global sort
order is required. For example, a global sort is a key step in computing medians or quantiles in
parallel; see Quantiling (page 89). The recipes in this chapter describe a simple parallel global
sort and variations on it to produce sorted serial output.

Simple global sort

To do a simple global sort, create a graph such as global_sort.mp (in the global-sort sandbox) by
following the steps summarized after the figure below and described in detail in the rest of this
chapter.

AB INITIO CONFIDENTIAL AND PROPRIETARY Global sort 57

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 1. Sample the data (page 59).
Because your graph will need to partition the input data by key range, you need to calculate
which values of the key field will divide the data into groups of roughly equal size. Calculating
these dividing points, or splitters, by reading all the records would be inefficient and
unnecessarily precise; it is better to take a random sample of the data and calculate the
splitters just for the sample.
2. Calculate and broadcast the splitter values for partitioning the data by key (page 59).
The FIND SPLITTERS component reads its entire input flow into memory, sorts it by key, and
outputs records that are equally spaced in the sorted order, thus dividing the input records as
evenly as possible. The number of output records is one less than the number of partitions,
which you specify as the num_partitions parameter. FIND SPLITTERS is designed to produce the
division of records that will yield the lowest possible skew when the splitters are used by the
PARTITION BY RANGE component.

58 Global sort AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The BROADCAST component writes a copy of the splitters to each of its output flow partitions ABOUT SAMPLE SIZE AND
and sends them to the split port of the PARTITION BY RANGE component. SKEW
3. Partition the data by the splitter values and sort each partition (page 60). If simply calculating a sample size of
about 10 MB is not acceptable, and
The PARTITION BY RANGE component uses the splitters to direct the records from the full input
you want to have more precise control
dataset to the appropriate partitions, with each partition getting approximately equal over the skew in the sorted partitions,
numbers of records with successively later keys in the sort order. The SORT component then choose a value for sample_size by
sorts the records in each partition to produce the final output. balancing these considerations:
z The sample_size parameter,
multiplied by the size of the
Sampling the data records in the input, should be
less than the amount of memory
To produce a relatively small random sample of the input data for use in determining the to be used by the FIND SPLITTERS
splitters: component. FIND SPLITTERS must
z Connect a SAMPLE component to the input file and set the following parameters: run in a serial layout (see the
next section) so that it can bring
sample_size (number of records) records from all partitions
max_skew (optional: default is 1) together. If it runs out of memory,
the graph will fail.
You will probably get acceptable results simply by choosing a sample_size that results in a
z The sample size should be large
sample of about 10 MB, and using the default max_skew value (1.0). However, if you want to enough to produce splitters that
take a more careful approach, follow the guidelines described in the sidebar. are reasonably close to the real
splitters and thus ensure a low
skew. A large sample is a more
Calculating and broadcasting the splitters accurate representation of the
input data; the larger the sample,
To calculate and broadcast the splitters: the smaller the skew.
For more information, see the
1. To sort the records according to a key specifier and find the ranges of key values that divide
SAMPLE component in Ab Initio Help.
the records approximately evenly into a specified number of partitions, connect a FIND
SPLITTERS component to the SAMPLE component and configure it as follows:

a. Set its key specifier to the field on which to sort. (In global_sort.mp, this is the id field.)

AB INITIO CONFIDENTIAL AND PROPRIETARY Global sort 59

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. Use the following command to set the num_partitions parameter to be dynamically
calculated at runtime using the value of $AI_MFS:
$(m_expand -n $AI_MFS/global_sort_in.dat)
The FIND SPLITTERS component needs to be given the degree of parallelism of the layout
it is partitioning to. However, you should not hardcode num_partitions, as you might
want it to be different in different environments (for example, development versus
production). It should also not be a separate parameter, because that would make it too
easy to choose one depth for $AI_MFS and another depth for this parameter.
c. Set the Interpretation of num_partitions to shell.
d. On the Layout tab of the Properties dialog, set the URL to $AI_SERIAL.
2. To spread the splitters back into the parallel layout of the rest of the graph, connect a
BROADCAST component to the FIND SPLITTERS component.

Partitioning by range and sorting the partitions

To divide the data into partitions according to the key field ranges calculated by the FIND
SPLITTERS component:
1. Add a PARTITION BY RANGE component and configure it as follows:
a. Set its key to the same field as the one used in the FIND SPLITTERS component, or use a
more specific key if necessary. (In global_sort.mp, this is the id field.)
b. Connect its split port to the BROADCAST component.
c. Connect its in port to the input file in the main graph.
2. To sort the newly partitioned data by id within each partition, attach a SORT component to
the PARTITION BY RANGE component and connect it to an output file.

60 Global sort AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

3. Configure the SORT component as follows: PRESORTED INPUT
a. Set the key to the same field as the one used in the FIND SPLITTERS and PARTITION BY If the input file is already sorted
RANGE components. (In global_sort.mp, this is the id field.) within each partition, but not
between partitions, the performance
b. Set max-core as desired or leave the default (100 MB). of global_sort.mp is poorer than you
might expect. This is because the
graph partially serializes: at first, all
Checking the results the partitions of the PARTITION BY
RANGE component send records to
To test the results of running global_sort.mp: the first partition of the SORT
z Run global_sort_check.mp. component. The node on which this
partition is running is working hard
If this graph finishes without error, the output is correctly sorted.
to sort, while the other nodes are
mostly idle. After a while, all the
partitions of the PARTITION BY RANGE
component switch to sending
records to the second partition of
the SORT component. Now the
second node is working hard, while
the others are idle. This pattern
continues, with all partitions of the
PARTITION BY RANGE component
switching at about the same time.
The overall performance of the
graph is only slightly better than if it
were running without any data
parallelism.

AB INITIO CONFIDENTIAL AND PROPRIETARY Global sort 61

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Producing serial output
If your goal is to produce a serial file, you might be tempted simply to add a CONCATENATE
component after the SORT, as in global_sort_serial_out.mp:

However, that will badly affect parallel performance, as successive partitions of the SORT
component will have to wait for the CONCATENATE to consume data from preceding partitions.
Instead, it would be better to address the problem ab initio freshly, from the beginning. Rather
than finding splitters and partitioning by range, you can simply partition in round-robin fashion
(evenly distributing the data), sort the partitioned data, and then collect the data using a MERGE
component (as in global_sort_serial_simple.mp).

NOTE: If you create these graphs by modifying global_sort.mp, do not forget to change the
URL for the output file to file:$AI_SERIAL/out_file.dat.

62 Global sort AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

AB INITIO CONFIDENTIAL AND PROPRIETARY Global sort 63

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Finding differences
Finding differences (also known as differencing or diffing) is useful in various situations:
z When constructing a graph to reimplement an existing process, you should test the
graph by running it on the same data as the old process, and comparing the results to
make sure they are the same.
z When updating or maintaining a graph, you should compare the outputs of the new
and old versions of the graph.
z When you have two versions of a dataset that were generated at different times, you
may want to compare them to see which records have changed. For example, you
might want to see which addresses have changed between two monthly versions of
your master customer file. You could use the results of the graph to determine which
rows need to be inserted, deleted, and updated in a database.
The recipe in this chapter describes how to create a graph that compares two datasets and finds
the differences between them. The graph outputs separate datasets containing the following:
z Added records
z Deleted records
z Updated records
z Changed fields

AB INITIO CONFIDENTIAL AND PROPRIETARY Finding differences 65

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The graph difference.mp, shown on page 67 and provided in the difference sandbox, finds the
differences between two input files. It makes the following assumptions about the input
datasets:
z Order and partitioning are unimportant. The two datasets are considered the same if
they contain the same records in any order and in any partition.
z Each record has one or more key fields, and each key is unique within the dataset.
For a generic version of a differencing graph in which transform functions and an output record
format are automatically generated from input parameters, see Automated generation of
record formats and transforms (page 79).

66 Finding differences AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

AB INITIO CONFIDENTIAL AND PROPRIETARY Finding differences 67

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The graph processes the input files and sends records to the outputs as described in the
following table:

OUTPUT RECORDS SENT TO OUTPUT

Trash Unchanged records (that is, records that are identical in both input files).

Different Data Records showing changes between the Old Data and New Data datasets. Each record
file includes the key field(s), and an old_field and new_field field for each non-key field:
z For non-key fields that have not changed, these fields are blank.
z For non-key fields that have changed, the old_ field and new_ field fields contain the
old and new values of the field.

Updates file Records from the New Data file that have keys matching those of records in the Old Data
file but have one or more differing non-key fields; in other words, these are updated
versions of existing records.

Adds file Records from the New Data file whose keys do not appear in a record in the Old Data file.

Deletes file Records from the Old Data file whose keys do not appear in a record in the New Data file.

Creating a differencing graph

To create a graph like difference.mp:
1. For the input files containing the old and the new data, make sure that all records with the
same key are in the same partition and are sorted within their partitions.
To do this, connect a PARTITION BY KEY AND SORT component to each input file, and set its key
appropriately.

68 Finding differences AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 For difference.mp, the record format of the input files is as follows:
record
date("YYYYMMDD") date_key;
string(8) str_key;
decimal(8) dec_field;
little endian integer(8) int_field;
string("\n") str_field;
end
The key is formed by the first two fields:
{date_key; str_key};
2. To determine which records have been added or deleted, and to collect the records that exist
in both the old and new datasets, use a JOIN component.
Connect the JOIN component to the two PARTITION BY KEY AND SORT components, name it
appropriately (in difference.mp, this is Check for Matching Keys), and configure it as follows:
a. Set the join_type parameter to Inner Join (matching records required on all inputs).
b. Set the key parameter to the same value as the key for the PARTITION BY KEY AND SORT
components you used in Step 1.
c. To collect records that have been deleted (records that are present in the old dataset
but not in the new one), connect the unused0 port to an output dataset, and name the
dataset appropriately (in difference.mp, this is the Deletes dataset).
d. To collect records that have been added (records that are present in the new dataset
but not in the old one), connect the unused1 port to an output dataset, and name the
dataset appropriately (in difference.mp, this is the Adds dataset).
If necessary, choose View > Optional Ports to show all optional ports. Then connect the
unused0 and unused1 ports to the output datasets, and deselect Optional Ports to hide
the unneeded ports.

AB INITIO CONFIDENTIAL AND PROPRIETARY Finding differences 69

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 e. For the out port, create an output record format that includes a field for both the old
and the new values of each field, so that you can later compare the values to see which
ones have changed. In the difference.mp graph, this is difference_both.dml:
record
date("YYYYMMDD") date_key;
string(8)str_key;
decimal('') old_dec_field;
decimal('') new_dec_field;
decimal('') old_int_field;
decimal('') new_int_field;
string("\n") old_str_field;
string("\n") new_str_field;
string(1) newline = "\n";
end;
f. To collect records that exist in both the old and the new datasets, edit the transform
parameter such that it writes all matching old and new fields from each input file to the
out port. The difference.mp graph does this by using the file combine.xfr, which
contains the following transform:
out :: join(old, new) =
begin
out.date_key :: old.date_key;
out.str_key :: old.str_key;
out.old_dec_field :: old.dec_field;
out.new_dec_field :: new.dec_field;
out.old_int_field :: old.int_field;
out.new_int_field :: new.int_field;
out.old_str_field :: old.str_field;
out.new_str_field :: new.str_field;
end;

70 Finding differences AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

3. To find the differences among records that exist in both the old and the new datasets, use a
REFORMAT component.
Connect the REFORMAT component to the JOIN from Step 2, name it appropriately (in
difference.mp, this is Split Out Records with Differences), and configure it as follows:
a. Set reject_threshold to Never abort.
b. Create a transform that compares the new and the old value for each field, sends
records that do not have differences to a TRASH component, and rejects records that do
have differences. (You will use the rejected records to generate two of the output
datasets.)
The difference.mp graph does this by using the transform diff_records.xfr:
out :: reformat(in) =
begin
out.date_key :: in.date_key;
out.str_key :: in.str_key;
out.dec_field :: if (in.old_dec_field == in.new_dec_field)
in.old_dec_field;
out.int_field :: if (in.old_int_field == in.new_int_field)
in.old_int_field;
out.str_field :: if (in.old_str_field == in.new_str_field)
in.old_str_field;
end;

NOTE: This transform works because the fields in the output record format (specified
in difference_input.dml) do not have default values. For fields with default values,
you would have to use something like the following to ensure that records with
differences would be sent to the reject port:
out.dec_field ::
if (in.old_dec_field == in.new_dec_field)
in.old_dec_field
else force_error("Old and new values are different");

AB INITIO CONFIDENTIAL AND PROPRIETARY Finding differences 71

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 4. Replicate the flow from the reject port of the Split Out Records with Differences component
(these are the records with differences).
5. Use one REFORMAT component to output a dataset containing the differences between the old
and the new data.
Name the REFORMAT and the output dataset appropriately and code the transform of the
REFORMAT as shown below. (In difference.mp, the REFORMAT is Leave Only Different Fields,
the output dataset is Different data, and the transform is diff_fields.xfr.)
a. Output the old and new values of the key fields.
b. Compare the old and new values of the other key fields. If they are different, output
each value. If they are the same, output a blank field.
out :: reformat(in) =
begin
out.date_key :: in.date_key;
out.str_key :: in.str_key;
out.old_dec_field :: if (in.old_dec_field != in.new_dec_field)
in.old_dec_field else "";
out.new_dec_field :: if (in.old_dec_field != in.new_dec_field)
in.new_dec_field else "";
out.old_int_field :: if (in.old_int_field != in.new_int_field)
in.old_int_field else "";
out.new_int_field :: if (in.old_int_field != in.new_int_field)
in.new_int_field else "";
out.old_str_field :: if (in.old_str_field != in.new_str_field)
in.old_str_field else "";
out.new_str_field :: if (in.old_str_field != in.new_str_field)
in.new_str_field else "";
end;
The above transform sends the following data to the output file Different Data:

72 Finding differences AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

6. Use another REFORMAT component to output a dataset containing the updated records. Name
the REFORMAT and the output dataset appropriately (in differences.mp, they are named
Output Updated Records and Updates) and code the transform of the REFORMAT to assign
the new value of each field to the updated records. In differences.mp, the transform is
diff_only_new_fields.xfr:
out :: reformat(in) =
begin
out.date_key :: in.date_key;
out.str_key :: in.str_key;
out.dec_field :: in.new_dec_field;
out.int_field :: in.new_int_field;
out.str_field :: in.new_str_field;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Finding differences 73

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Variations on simple differencing
The graph difference.mp makes certain assumptions about your goals in comparing two
datasets. If your goals are different, you can change the graph as needed. For example, you can:
z Omit the Different Data output dataset. Do this if you merely want to find the set of
rows to be added, dropped, and updated in a dataset. In this case you can also
eliminate the REPLICATE and Leave Only Different Fields components, and connect
Output Updated Records directly to Split Out Records with Differences via a straight
flow.
z Use an Output file component instead of a TRASH component. Do this if you want to
keep unchanged records rather than discarding them. Simply attach the output dataset
to the Split Out Records with Differences component.

More complex differencing: Dealing with duplicate key values

The graph difference.mp works only for data with unique key values. To use a similar graph to
find differences between datasets when one or both of them have duplicate keys, you must add
logic to specify which new records correspond to old records. One way to do this is by using a
SCAN component to add a uniquely numbered field to each record in a key group, thereby
creating a two-field key for each record. For details on how to do this, see Making unique keys
(page 21).

74 Finding differences AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Advanced differencing: Equality checks
Data-checking graphs must be able to process all records in a dataset without causing errors
that stop the graph. This means that they must handle potentially corrupted data, no matter
how mangled the contents of a field, and be able to detect the beginning and end of each
record. To ensure that your graphs can do this, follow the guidelines in this section.

Safe equality checks

The rules for safe comparison depend on the types of the values being compared.

string and integer types Since all combinations of characters are valid strings, and all sequences
of bytes are valid binary integers, you can safely use the non-equal operator (!=) to compare
string and integer types. The comparison expression for a field named f is simply:
in0.f != in1.f

decimal, date, and datetime types If you think the quality of your data is questionable, check the
validity of decimal, date, and datetime values before comparing them. Do not simply cast them
to strings before comparing them, as values that are equal in their original types may have
different representations as strings. For example, 5 and 5.0 are equal as decimals, and
22APR1962 and 22apr1962 are equal as dates, but neither of these pairs are equivalent as
strings. Then do one of the following:
z If they are valid, compare them as their actual types.
z If they are invalid, cast them to strings of the appropriate length, and then compare
them as strings.

AB INITIO CONFIDENTIAL AND PROPRIETARY Finding differences 75

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 For a field f of type decimal(","), use the following comparison expression:
if (is_valid(in0.f) && is_valid(in1.f))
in0.f != in1.f
else
(string(","))in0.f != (string(","))in1.f

real types You can safely compare the values of real types with the non-equal operator, in the
sense that this operation never produces an error. The expression in0.f != in1.f, for example,
never rejects a record.
Note, however, that the IEEE representation accommodates a value that is not a number (NaN).
Such values always compare as unequal. If you think your dataset may contain NaNs, you
should take special care to treat all NaNs as equal to one another by using the math_isnan
function in a comparison expression like this:
if (math_isnan(in0.f) && math_isnan(in1.f))
0
else
in0.f != in1.f

void types You cannot directly compare the values of void types. Instead, reinterpret them as
strings and then compare the strings. To compare two fields of type void(100), use an expression
like this:
reinterpret_as(string(100), in0.f)
!= reinterpret_as(string(100), in1.f)

76 Finding differences AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

vector types You must compare values of type vector element by element, using the
appropriate rule for the vectors element type. You may want to define a function like the one
shown below for comparing vectors. It first declares variables to hold the result of the
comparison and the iterator for the for loop. Then it compares the lengths of the vectors, and
finally the individual elements.
out::vectors_unequal(in0, in1) =
begin
let int result = 0;
let int i;
if (length_of(in0) != length_of(in1))
result = 1;
else
for (i, i < length_of(in0) and result == 0)
if (in0[i] != in1[i])
result = 1;
out :: result;
end;

This function works for vectors of data types that can be compared using the non-equal
operator (!=). For vectors of other types such as reals, subrecords, and vectors you may
need to replace this expression:
in0[i] != in1[i]

with a better test, as described elsewhere in this section.

Then use a comparison expression like this:
vectors_unequal(in0.f, in1.f)

record types Compare the values of record types field by field, using the appropriate rules for
the type.

union types Compare the values of union types by comparing their largest members.

AB INITIO CONFIDENTIAL AND PROPRIETARY Finding differences 77

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Fuzzy equality checks
Because floating-point numbers are an imprecise representation of real numbers, some
operations can produce surprising results that do not always agree with the usual rules of
arithmetic. For example, the result of adding floating-point numbers depends on the order in
which the numbers are added: X+Y+Z may be slightly different from Z+X+Y.
Typically the differences are small enough to ignore, and using the floating-point representation
is acceptable. However, when comparing the output of two processes that use floating-point
numbers, you should be prepared to accept a certain level of difference: the output of the two
processes may differ slightly, without either being wrong.
To compare records produced by processes that use floating-point arithmetic, replace this
comparison expression:
in0.f != in1.f

with this:
math_abs(in0.f - in1.f) > 1e-10
Here the tolerance has been arbitrarily chosen to be .00000000001 (or 10-10). Setting the
tolerance for a comparison requires consideration of business rules (what is too small to
matter?) and the details of floating-point arithmetic (how close can you expect the values to
be?). You may also find it useful to create a function that defines the tolerance for a large
collection of graphs or projects.

78 Finding differences AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Automated generation of record formats
and transforms
The recipes in this chapter describe metaprogramming techniques for automating the creation
of record formats and transforms. They present a parameterized metadata generation graph
(make_difference_metadata.mp, in the gen-diff sandbox) that generates transforms and a
record format for finding and showing the differences between two input datasets. The
transforms and record format it generates serve the same purpose as those in difference.mp
(described in Finding differences on page 65) but are built automatically from a specified input
record format. The outputs of make_difference_metadata.mp are used by
generic_difference.mp.

Generated record formats and transforms can be useful in all sorts of other cases. For example:
z Data validation frequently requires applying one or more tests to all fields in each
record and attaching to each unacceptable field an identification of which test(s) failed.
To automate the generation of a transform and an output record format for such
validation, you could create a graph that produces them from an input record format
and a list of rules.
z Developers who receive field names and types in formats other than DML files (for
example, data dictionaries in the form of spreadsheets) could create a graph to convert
such information into DML files.

AB INITIO CONFIDENTIAL AND PROPRIETARY Automated generation of record formats and transforms 79

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Generic differencing graph
Following is a review of the most important components in difference.mp (presented in Finding
differences on page 65) and generic_difference.mp (page 81), and a summary of how the
metadata generation graph (page 84) creates generic transforms like the ones used in those
components.

80 Automated generation of record formats and transforms AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

AB INITIO CONFIDENTIAL AND PROPRIETARY Automated generation of record formats and transforms 81

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The bulk of the metadata generation graphs work is done by its REFORMAT component, which
also contains a transform that creates the DML file needed for the output of the differencing
graph. (For details, see Finding differences on page 65.)
z Join: Check for Matching Keys Determines which records have been deleted or
added, and collects the records that exist in both the old and the new datasets. For its
out port, it uses a record format that includes a field for both the old and new values,
each prefixed by old_ or new_ (except key fields and ignored fields).
The REFORMAT component in the metadata generation graph creates a transform that
checks for key fields and ignored fields and leaves their names unchanged, while for all
other fields, it adds the prefix old_ or new_, based on whether the record came from
the old or new dataset.
z Reformat: Split Out Records with Differences Finds records that are different
between the old and new datasets by comparing the old and new value for each field.
The REFORMAT component in the metadata generation graph creates a transform that
checks for key fields and ignored fields and leaves their names unchanged, while for all
other fields it checks whether the old value and the new value are the same, and if they
are, it assigns the old_ value to each out field. This causes records with differences to
be sent to the reject port.
z Reformat: Leave Only Different Fields Outputs a dataset containing the differences
between the old and the new data. For records that are different, each value is in a field
whose name is appropriately prefixed with old_ or new_. For records that are the same,
those fields are blank. Key fields and the newline field are left unchanged, while for any
other ignored fields, an empty string is output.
The REFORMAT component in the metadata generation graph creates a transform that
first checks for key fields and ignored fields. For key fields and newline fields, it assigns
the in value to the out value, while for any other ignored fields, it outputs an empty
string. For all other fields, it adds the prefixes old_ and new_ to the field names

82 Automated generation of record formats and transforms AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 appropriately, compares the old and the new values, and assigns the appropriate strings
to the output fields.
z Reformat: Output Updated Records Outputs a dataset containing the updated
records, prefixing each out field (except key fields and ignored fields) with new_.
The REFORMAT component in the metadata generation graph creates a transform that
checks for key fields and ignored fields, leaving their names unchanged. For records
that have changed, it outputs a dataset containing the updated records, assigning the
new value to each field and adding the prefix new_ to the field name.

AB INITIO CONFIDENTIAL AND PROPRIETARY Automated generation of record formats and transforms 83

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Metadata generation graph

To parse the passed-in lists of key fields and ignored fields (including the newline field) into
vectors of strings to be stored in the Special Fields lookup file, the metadata generation graph
uses two NORMALIZE components that call the user-defined function parse_field_list. A GATHER
component then combines the two flows and outputs them to the lookup file.

84 Automated generation of record formats and transforms AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

DML metaprogramming record_info, length_of, AND
RELATED TYPES
The transforms in the REFORMAT component of make_difference_metadata.mp make frequent The DML function record_info returns
use of several DML metaprogramming functions and the built-in types that support them. The a vector of the attributes of each field
functions produce output strings in the form of DML record formats (type definitions) and from the passed-in record
information. The built-in DML type
transform rules that can then be used in other graphs. The following sections describe how the
dml_field_info_vec contains a
REFORMAT transforms use string manipulation and more complex logic to produce the record
dml_field_info record for each field,
formats and transforms. as shown below. (For more
information, see Ab Initio Help.)
type dml_field_info =
Using DML to generate record formats record
string(integer(4))
name;
The first transform in the REFORMAT component takes a string containing information about the string(integer(4))
dml_type;
record format of the input files (Old Data and New Data) and generates from it a DML file with string(integer(4))
default;
fields for both the old and the new values of each record as follows: decimal(1)
nullable;
1. Declares variables to hold the attributes of each field in a record (info vector), the number of decimal(1)
nullflag;
elements in info, the differences between fields, the name and DML type of each field in the string(integer(4))
condition;
record, and a loop counter: string(integer(4))
out :: reformat(in) = form;
end;
begin
let dml_field_info_vec info = record_info(in); The length_of function, when passed
let int n_fields = length_of(info); a value of type dml_field_info_vec,
let dml_field_info_vec diff_info = [vector]; returns the number of elements in the
let string('') name; vector.
let string('') dml_type;
let int i;
2. Loops through each element in the dml_field_info_vec and does the following:
a. Gets the name of each field in the vector.
b. Sets dml_type appropriately.

AB INITIO CONFIDENTIAL AND PROPRIETARY Automated generation of record formats and transforms 85

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

make_field AND add_fields c. Checks whether name is in the Special Fields lookup file, which contains
The DML function make_field creates (parameterized) key fields and ignored fields.
a dml_field_info record from strings If it is, the transform simply appends the record info for that field to the diff_info
containing field information. vector.
The DML function add_fields adds a
If it is not, the transform creates a diff_info vector of differences by adding the
new field definition to the text of a
strings new_ and old_ to the name field, using the dml_type that was set earlier,
record or union definition. In the
transforms of the REFORMAT and taking the default and nullflag attributes from the dml_field_info_vec
component of make_difference_ unchanged.
metadata.mp, it adds the content of for (i, i < n_fields)
the diff_info vector to a record begin
definition. (For more information, name = info[i].name;
see Ab Initio Help.)
if (info[i].form member [vector 'integer', 'real',
'decimal'])
dml_type = 'decimal("")';
else
dml_type = info[i].dml_type;

if (lookup_match("Lookup: Special Fields", name))

diff_info = vector_append(diff_info, info[i]);
else
begin
diff_info = vector_append(diff_info,
make_field(string_concat('new_', name),
dml_type, info[i].default, '',
info[i].nullflag));
diff_info = vector_append(diff_info,
make_field(string_concat('old_', name),
dml_type, info[i].default, '',
info[i].nullflag));
end
end

86 Automated generation of record formats and transforms AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 NOTE: As the assignment of dml_type above shows, you can write code that does dml_rule_vec, make_rule,
more than simple string manipulation to generate a record format; here the type is AND add_rules
set to decimal if the form attribute of the input dml_field_info record is integer, real, The DML type dml_rule_vec is a
or decimal. vector of DML rule definitions of the
3. Calls add_fields to add the vector of differences to the type definition: following form:
type dml_rule =
out :: add_fields('record end', diff_info); record
end; string(integer(4))
lhs;
string(integer(4))
rhs;
Using DML to generate transforms decimal(',')
priority;
end;
The last four transforms in the REFORMAT component use the dml_field_info_vec returned by where lhs is the left-hand side of
record_info in a fashion very similar to the first one (transform0), but they generate transforms the rule, rhs is the right-hand side,
rather than a record format. and priority is the rules priority.
Three of these transforms (transform1, transform2, and transform4) differ from transform0 in
the following ways:
z They declare a variable of type dml_rule_vec to hold transform rule definitions (rather
than a variable of type dml_field_info_vec to hold field attributes):
let dml_rule_vec rules = [vector];
z They call make_rule to create a transform rule definition (rather than make_field to
create a dml_field_info record):
rules = vector_append(rules, make_rule(string_concat(...
z They call add_rules to add a vector of rule definitions to the transform definition
(rather than add_fields to add a vector of differences to a type definition):
out :: add_rules('out :: ...= begin end;', rules);

For more information on these functions, see Ab Initio Help.

AB INITIO CONFIDENTIAL AND PROPRIETARY Automated generation of record formats and transforms 87

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The three REFORMAT transforms described above use very simple logic to handle fields in the
Special Fields lookup file whose field_kind is key or ignored. They call lookup_match for each
field in the dml_field_info_vec to check whether the field name exists in the lookup file. If it
does, each transform takes the same action in all cases, regardless of the field_kind attribute or
the name of the field.
However, transform3 demonstrates that you can automate more complex operations: it calls
lookup to get the field_kind attribute for each field from the Special Fields lookup table and to
check whether the fields name is newline. For fields whose field_kind is key or whose name is
newline, it adds the prefix in, while for any other ignored fields, it outputs a blank string.
if (lookup_match("Lookup: Special Fields", name))
rules = vector_append(rules,
make_rule(string_concat('out.', name),
if (lookup("Lookup: Special Fields",
name).field_kind == 'key' or
name == 'newline')
string_concat('in.', name)
else
'""'));

88 Automated generation of record formats and transforms AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Quantiling
Quantiling can be defined as determining the ranges of a value that most evenly divide a set of
records into a specified number of groups, and assigning each record to the appropriate group.
Business applications frequently use quantiling to rank customers or otherwise divide them into
groups for further statistical analysis. Deciling is a specific form of quantiling in which the
records are divided into 10 groups.
The recipes in this chapter describe how to create graphs that do the following:
z Simple serial deciling Divide records evenly into 10 groups (page 90).
z Fix decile boundaries Ensure that all records with the same scored value are
assigned to the same group (page 92).
z Parallel deciling Divide records that are distributed across partitions into 10 groups
(page 95).
z Parameterized parallel quantiling Allow users to customize a parallel quantiling
graph by specifying values for various parameters (page 103).
The input file used for these graphs consists of 1,000 records that have a scored_value field for
calculating the quantiles.

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 89

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Simple serial deciling
This recipe describes how to create a serial graph (quantile1.mp in the quantile sandbox) that
uses the scored_value field of the input records to divide them into 10 groups of roughly equal
size.

NOTE: This recipe demonstrates one of the key ideas behind the basics of quantiling but is
flawed, as described later. The recipe Fixing decile boundaries (page 92) addresses the flaw.
To create a serial graph like quantile1.mp:
1. Connect a ROLLUP component to your input file and use it to count the records, as follows:
a. Set key to the empty key: {}

90 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. In the ROLLUPs transform, call count(1) to count the records: EMPTY KEY ON JOIN
out::rollup(in) = An empty key on a JOIN causes all
begin records on each port to be matched
out.count:: count(1); with all records on all other ports. In
end; this case it is a simple way to match
2. In a separate flow, add a SORT component that sorts the input file by scored_value. up the single output record from the
ROLLUP with every record from the
3. Connect the ROLLUP and the SORT to a JOIN component, and do the following in the JOIN:
SORT. Another way to do this would be
a. Set key to the empty key: {} to match all records from the ROLLUP (in this case only one) to write the output of the ROLLUP into
with all records from the SORT. a LOOKUP FILE and read it with a
REFORMAT.
b. Set the parameter sorted_input to In-memory: Inputs need not be sorted.
c. Calculate the decile for each input record as shown in the following code, and output
each scored_value with its decile:
let integer(8) record_number = 0;

out::join(in0, in1) =
begin
record_number = record_number + 1;

out.* :: in0.*;
out.decile :: ((record_number * 10 - 1) / in1.count) + 1;
end;
This code assigns the correct decile to each record in sequence: (record_number * 10
1) results in a number from 9 (for the first record) to just under 10 times the number of
records (for the last record). That quantity divided by the number of records yields a
value between 0 (for the early records) and 9 (for the last records). To label the deciles 1
through 10, the code adds 1 to that value.
4. Run the graph and examine the output data to find the flaw in this graph:

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 91

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The records are divided into exactly 10 groups of 100 each, but in at least one case, two records
with the same scored_value fall into two different deciles: Records 400 and 401 both have
scored_value 404, but the first one is marked as belonging to decile 4 and second one as
belonging to decile 5. To fix this problem, follow the steps in the next section.

Fixing decile boundaries

To ensure that all records with the same scored_value are assigned to the same decile, the graph
needs to remember the value of decile for records that share the same scored_value. The best
way to do this is to use a SCAN component as shown in the next figure (quantile2.mp). If you
code the SCAN components functions as described after the figure, they will behave as follows:

92 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 z initialize Is called at the beginning of each group of records that have the same key
(scored_value), and stores (in its temporary_type variable) the decile for the first record
in each group
z scan Remembers the decile value as it processes each record in the group
z finalize Puts the decile value in the decile field of every record in the group

By assigning the same decile value to all records in the same key group (specifically, the decile
value of the first record in the group), the SCAN fixes any bad boundaries caused by the JOIN.

To use a SCAN to fix decile boundaries:

1. In a copy of the graph described in Simple serial deciling (page 90) (quantile1.mp), insert and
connect a SCAN component between the JOIN and Deciled Records.

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 93

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 2. Configure the SCAN component as follows:
a. Specify scored_value as the key.
b. Set the sorted_input parameter to Input must sorted or grouped.
c. On the Ports tab, select the Propagate through checkbox.
3. Double-click the SCAN component and allow the GDE to generate a default transform.
4. In the Record Format Editor or in Text View, define the temporary_type as integer(4) decile:
type temporary_type =
record
integer(4) decile;
end
5. In the Transform Editor or in Text View, do the following:
a. In the initialize function, set temp.decile to in.decile.
b. In the scan function, set temp.decile to temp.decile.
c. In the finalize function, set out.decile to temp.decile and out.scored_value to
in.scored_value:
temp::initialize(in) =
begin
temp.decile :: in.decile;
end;

temp::scan(temp, in) =
begin
temp.decile :: temp.decile;
end;

out::finalize(temp, in) =
begin
out.decile :: temp.decile;
out.* :: in.*;
end;

94 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

6. Run the graph and examine the output data:

Parallel deciling
The above descriptions of deciling are sufficient for serial data. However, if you are processing
large amounts of data in parallel, the records that belong in each decile will probably be
distributed across partitions, so you must take the steps summarized on page 96 to find and fix
the decile boundaries in parallel (see quantile3.mp).

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 95

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 To find and fix the decile boundaries in parallel:
1. Sort the data globally, so that the data is sorted within each partition and across partitions,
with the smallest values in the first partition and the largest values in the last. (For details on
how to do this, see Global sort on page 57.)
2. Replicate the resulting data so that you can use it for further processing and in the final JOIN.
3. Get the total record counts and starting record counts for each partition so that when you
calculate and assign the deciles, you know the relative position of any given record in the entire
set (page 97).

96 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

4. Calculate the splitters (the ending decile values) and fix the boundaries that is, ensure that
all records with the same scored_value will be assigned to the same decile, even though they
could be in different partitions (page 101).
5. Prepare the decile splitters for the final JOIN concatenate them, create a vector of their
values, and broadcast the vector (page 102).
6. Join the globally sorted data with the vector of splitters to assign the deciles and output the
data (page 102).

Getting the total record counts and starting record counts for each partition
To calculate the deciles properly, you must calculate the total count of records across all
partitions, as well as the starting count for each partition (that is, the total count of records in the
earlier partitions). You can do this by creating a subgraph similar to Get Counts in quantile3.mp
as described after the following figure, and connecting it to one of the out ports of the REPLICATE.

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 97

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

98 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

To create a subgraph similar to Get Counts in quantile3.mp:
1. Calculate the total count for each partition by using a ROLLUP component with its key set to
the empty key: {}
2. Reduce the counts from each partition to a single, ordered, serial flow and make a copy of it by
using an INTERLEAVE and a REPLICATE.
3. Compute a grand total of the counts from the partitions by using a ROLLUP with its key set to
the empty key: {} and broadcast it to all partitions.
4. Compute the starting count for each partition and send the counts to their proper partitions by
using a SCAN and then a PARTITION BY ROUND-ROBIN. Code the SCAN transform as follows (see
the code in Step c below):
a. Initialize its temporary_type to zero.
b. Read the stream of total counts for each partition (from the INTERLEAVE) and add each
one to the temporary_type to get the ending count of each partition.
c. Subtract each total count from the temporary_type to get the starting count of each
partition.
type temporary_type = decimal(12);

temp::initialize(in) =
begin
temp :: 0;
end;

temp::scan(temp, in) =
begin
temp :: temp + in.count;
end;

out::finalize(temp, in) =
begin
out.starting_count :: temp - in.count;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 99

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

USING WATCHERS This transform calculates a running count minus the current count, and yields the
Watchers are saved with a graph following values:
unless you explicitly delete them by
right-clicking the watcher and in.count temp out.starting_count
deselecting Watcher on Flow, or by
42 42 0
choosing Debugger > Remove All
Watchers. To see the watchers 221 263 42
saved with a graph, enable the
95 358 263
debugger by choosing Debugger >
Enable Debugger or clicking the 194 552 358
Enable icon on the Debugging
toolbar. When the debugger is 448 1000 552
disabled, the watchers will not 5. Optionally (if you would like to be able to view the data generated by the INTERLEAVE and SCAN
affect execution of the graph in
components), choose Debugger > Enable Debugger, right-click the out flows for those
any way.
components, and click Watcher on Flow. After running the graph, right-click each Watcher
icon and choose View Data to see the values.
INTERLEAVE watcher values for count:
42
221
95
194
448
SCAN watcher values for count (starting count per partition):
0
42
263
358
552

100 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Calculating the splitters and fixing the boundaries across partitions
To calculate the deciles to be used as splitters in the final JOIN, you can use the same logic as in
Simple serial deciling (page 90), with minor modifications. Instead of simply using
record_number, you must add in the starting count for each partition (in1.count +
record_number). Note also that the total count is now on in2.

A more complex task is fixing the boundaries to ensure that all records with the same
scored_value will be assigned to the same decile, even though they could be in different
partitions. You must write a JOIN transform that outputs a record only when it is the last record
assigned a given decile value. The transform shown below does this by comparing the decile of
the current record to that of the next, and putting out a record only when they differ. Here the
inputs are the following:
z in0 is the partitioned, sorted records (from the Global Sort subgraph).
z in1 is the starting counts for each partition (from the Get Counts subgraph).
z in2 is the total record count (from the Get Counts subgraph).
let integer(8) record_number =0;
let integer(4) this_decile = -1;
let integer(4) next_decile =-1;

out::join(in0, in1, in2) =

begin
record_number = record_number + 1;
this_decile = if (record_number != 1)
next_decile
else
(((in1.starting_count + record_number) * 10 - 1)
/ in2.count) + 1;
next_decile = (((in1.starting_count + record_number) * 10)
/ in2.count) + 1;

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 101

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 out.* :: in0.*;
out.decile :: if (this_decile != next_decile) this_decile;
end;

Preparing the decile splitters for the final JOIN

To present the set of decile splitters to the final JOIN as a vector of values, pass the calculated
values to the following components:
z CONCATENATE (to collect them in order)

z DENORMALIZE SORTED with its key set to the empty key: { } (to build a vector out of
them)
z BROADCAST (to send the vector to each partition of the JOIN)

Assigning the deciles and outputting the data

Finally, you need to assign a decile to each of the original 1,000 records and output the records
with their deciles. To do this, add a JOIN component and attach the following to its in ports:
z in0 a REPLICATE component containing the globally sorted records
z in1 a BROADCAST component containing the vector of splitter values for assigning the
deciles

102 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Write a transform such as the following to use the scored_value in each record to determine the
corresponding decile:
let integer(4) current_decile = 0;
let decimal(12) next_decile_boundary = 0;

out::join(in0, in1) =
begin
if (current_decile == 0)
begin
next_decile_boundary = in1.splitters[current_decile];
current_decile = 1;
end
while (in0.scored_value > next_decile_boundary)
begin
next_decile_boundary = in1.splitters[current_decile];
current_decile = current_decile + 1;
end

out.decile :: current_decile;
out.* :: in0.*;
end;

Parameterized parallel quantiling

This recipe describes how to create a generic version of quantile3.mp that uses parameters
instead of hardcoded values for the following:
z The name of the field used to calculate the quantiles
z The data type of the field used to calculate the quantiles
z The name of the field to write the quantiles to
z The number of quantiles to use

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 103

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

WHY YOU SHOULD CONNECT To create this parameterized version of the deciling graph:
FLOWS DIRECTLY TO INPUT NOTE: These steps are described in detail in subsequent sections.
FILES INSTEAD OF USING
REPLICATE 1. Create a quantiling subgraph from the original graph and insert it into a test graph.
When multiple components are 2. In the subgraph, add and export parameters for the values you do not want to be hardcoded
connected to an INPUT FILE, they each (page 106).
read the data independently, at their
3. In the subgraph, replace previously hardcoded values with parameters, and change their
own pace. When multiple components
are connected to the output of a interpretation appropriately (page 107).
REPLICATE, there are two potential 4. Test your quantiling subgraph (page 109).
problems:
z The downstream components
may be slowed to the processing Creating a quantiling subgraph and inserting it into a test graph
rate of the slowest component
among them. To create a quantiling subgraph from the original graph and insert it into a test graph:
z If the downstream components
1. Save copies of quantile3.mp under two new names such as the following:
send their outputs, directly or
indirectly, to a component in my-Subgraph_Parallel_Quantile.mp To become the quantiling subgraph
common, deadlock flow buffering my-quantile4.mp To be the test graph for the quantiling subgraph
will be introduced, possibly
causing data to spill to disk. 2. Delete the input and output files and make subgraph inputs and outputs from what were
Here the PARTITION BY RANGE previously their flows, as follows:
component in the Global Sort a. Drag two flows from the inputs of the Global Sort subgraph component to the left edge
subgraph cannot start partitioning the of the graph.
records until it reads the splitters
This allows the user of the subgraph to replicate the data explicitly if it is the result of
from the BROADCAST component. If
you use a REPLICATE to pass the input some earlier processing or to hook the two inputs directly to an INPUT FILE, which is more
data, the PARTITION BY RANGE will have efficient (see the sidebar).
to buffer the data while waiting for the b. Drag the flow from the output port of the JOIN to the right edge of the graph.
splitters.
3. Save the graph, ignoring the compilation error message.
4. Switch to my-quantile4.mp.

104 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

5. Delete everything except the input and output files. LINKED SUBGRAPHS
6. Save the graph, ignoring the compilation error message. The subgraph you inserted in Step 9
is a linked subgraph, as you can see
7. Choose Insert > Program Component and browse to the mp directory.
from its (read-only) legend when you
NOTE: In the Open dialog, make sure the Host button is selected so that you are browsing in open it from within my-quantile4.mp.
Host mode (not in Local mode). This means that the enclosing graph
contains a local copy with a link, or
8. Select my-Subgraph_Parallel_Quantile.mp, add $AI_MP/ before the filename (to ensure that it pointer, to the location of the original
is included with a sandbox path rather than hardcoded), and click Open. subgraph (in this case in $AI_MP).
9. Attach the flows from the subgraph to the input and output files. This location is displayed after
Source: on the Description tab of the
The graph should now look similar to quantile4.mp:
Properties dialog for the subgraph.
To edit a linked subgraph:
1. Browse to the location of the
subgraph, or right-click the copy
in the containing graph and
choose Component Source.
2. Edit and save the subgraph.
Changes you make to the original
subgraph are not automatically
updated in graphs that use them. This
prevents errors in the subgraph from
10.Save my-quantile4.mp. breaking every graph that uses it.

This graph should run, since you have not made any changes other than converting part of it to To update a containing graph with
changes made to a linked subgraph:
a subgraph.
z In the containing graph, right-
click the linked subgraph and
choose Update. (You can select
multiple components and
update all of them.)

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 105

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Adding and exporting parameters
The current version of the shared subgraph contains hardcoded values that reflect the dataset
used to test the original deciling graph.
To change the hardcoded values into parameters:
1. Open my-Subgraph_Parallel_Quantile.mp (not the linked copy of the subgraph contained in
the main graph) and choose Edit > Properties.
2. Create the four parameters listed below. To do this, click Create on the Parameters tab of the
Properties dialog, enter the names, and click OK. (The defaults for Type, Scope, and Kind are
acceptable.)
SCORED_VALUE To represent the name of the field used to calculate the quantiles.
(In the test data provided in the sandbox, this is scored_value.)
SCORED_VALUE_TYPE To represent the data type of the field used to calculate the
quantiles. (In the test data provided in the sandbox, this is decimal(12).)
QUANTILE_FIELD_NAME To represent the field that the quantile values will be
written to. (In the test data provided in the sandbox, this is decile.)
MAX_QUANTILES To represent the number of quantiles into which the data is to be
divided. (In the test data provided in the sandbox, this is 10.)
3. To export all parameters (except key) from the Global Sort subgraph so that they will inherit
their values from the containing graph, my-Subgraph_Parallel_Quantile.mp, do the following
for each parameter:
a. On the Parameters tab of the Properties dialog for the Global Sort subgraph, select the
parameter and click Export.
b. In the Exported Name box of the Export as Parameter of graph dialog, type the
appropriate name from the list below:

106 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 UNEXPORTED PARAMETER EXPORTED PARAMETER

serial_layout SERIAL_LAYOUT

parallel_layout PARALLEL_LAYOUT

SAMPLE_SIZE SAMPLE_SIZE

MAX_CORE MAX_CORE

The my-Subgraph_Parallel_Quantile.mp subgraph should now have eight parameters.

Replacing hardcoded values with parameters

To make the my-Subgraph_Parallel_Quantile.mp subgraph use the four parameters you created
in Step 2 of Adding and exporting parameters above (rather than the hardcoded values), do the
following:
1. Replace the hardcoded values as follows:
a. Choose Edit > Parameters.
b. On the left pane of the Edit Parameters dialog, click the plus signs to expand the
component listings for my-Subgraph_Parallel_Quantile, Subgraph_Get_Counts, and
Subgraph_Global_Sort.

NOTE: The grid in the Edit Parameters dialog allows global, graph-wide search and
replace operations, but only on items that are open in the view.

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 107

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 2. Choose Edit > Replace, click Match Case (to avoid making replacements in the parameters you
just defined), and make the global replacements shown in the table.
Start each search by selecting the top-level icon in the browser to ensure that no occurrences
are missed.

GLOBALLY REPLACE THIS WITH THIS

scored_value ${SCORED_VALUE}

decimal(12) ${SCORED_VALUE_TYPE}

3. Choose Edit > Replace, click Match Case (again to avoid making replacements in the
parameters you just defined), and where the values shown in the table below are values that you
want to parameterize, replace them as shown:
Do not replace the value 10 where it appears in the setting for max_core.
Do replace decile when it occurs in out_metadata; these are record formats used by the
components.
Do not replace decile in instances of [Same as], this_decile, or next_decile.

WHERE APPROPRIATE, REPLACE THIS WITH THIS

10 ${MAX_QUANTILES}

decile ${QUANTILE_FIELD_NAME}

4. For all the above replacements that occurred in a transform or record format, change the
interpretation from constant to ${} substitution as follows:
a. Choose Edit > Parameters.
b. On the left pane of the Edit Parameters dialog, click the plus signs to expand the
component listings for my-Subgraph_Parallel_Quantile, subgraph Get_Counts, and
subgraph Global_Sort.
c. Choose Edit > Find and search for ${ .

108 Quantiling AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 d. Wherever you find ${ in a transform or record format (for record formats, the Name in
the middle panel of the Edit Parameters dialog will end with _metadata), change the
interpretation by expanding the Advanced Attributes list in the right pane of the Edit
Parameters dialog (if necessary) and selecting ${} substitution instead of constant.
e. Save your changes.
5. Save your shared subgraph, ignoring the compilation error message (displayed because the
graph has no inputs and therefore cannot propagate record formats and layouts).

Testing your quantiling subgraph

To test the quantiling subgraph:
1. Switch to your test graph (my-quantile4.mp).
2. Right-click the my-Subgraph_Parallel_Quantile subgraph and click Update.
A yellow to-do cue appears because the four newly defined parameters have no values.
3. Provide the following values:

FOR THIS PARAMETER SPECIFY THIS VALUE

${SCORED_VALUE} scored_value

${SCORED_VALUE_TYPE} decimal(12)

${QUANTILE_FIELD_NAME decile

${MAX_QUANTILES} 10

4. Run the test graph.

AB INITIO CONFIDENTIAL AND PROPRIETARY Quantiling 109

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Auditing
In many situations it is important to record and compare the number of records flowing into and
out of a graph. In an accounting system, for example, you may need to ensure that a report
contains information on all accounts.
The two recipes in this chapter show how to develop graphs that meet auditing requirements for
financial applications. They present two different approaches to auditing the same sample
application:
z Using a tracking summary file and a separate graph to ensure that no records have
been lost between the inputs and outputs.
z Using log flows to ensure that no records have been lost and that the financial totals
are unchanged between the inputs and outputs

Sample application to audit

The following figure shows the graph to be audited (audit1.mp in the audit sandbox). It is a
simplified example of a banking application that creates a summary of the balances, debits,
credits, and customer information for each account:

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 111

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

112 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Simple auditing: Counting records TRACKING DATA
Ab Initio software provides tracking
This recipe shows how to audit a graphs record counts to make sure that no account records data in four ways:
have been lost between the inputs and outputs. z Numbers on the flows in the GDE
z Text sent to standard output
To audit a graphs record counts: when you run a graph from the
1. Write tracking information to a summary file and examine the file to identify the flow(s) you shell
will need for your audit (see the next section). z Values stored in the EME
z Tracking summary file
2. Create a separate graph that uses a ROLLUP to compare the input and output record counts
The numbers in the first two locations
from the summary file (page 115).
are not readily parsed, and not
everyone uses the EME to store
tracking results. This recipe describes
Writing tracking information to a summary file and finding the flows needed how to use the summary file.
for the audit
By default, the GDE uses the settings on the Tracking tab of the Run > Settings dialog to set the
configuration variable AB_REPORT, which controls the ways a graph's tracking information is
reported during a run. However, you can augment the AB_REPORT setting with options that are
not visible in that dialog. To do this, choose Edit > Script > Start and add a line such as the
following to the start script of the graph:
export AB_REPORT="$AB_REPORT other_settings"

By adding a setting of the form summary=path, you can tell the graph to write the tracking
information to the file specified in path, in a summary format that is easy to read from
another graph. (For details, see AB_REPORT and Generating a summary file in Ab Initio Help.)
The format of the summary file is specified in $AB_HOME/include/summary.dml, which
contains a record format with conditional fields: the presence or absence of the fields in a
specific summary file is determined by the kind of object (flow, component, job-start, or phase-
end) being reported on. After running the graph, you can examine the summary file to find the
flow(s) that are relevant for your audit.

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 113

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

FLOWS IN SUMMARY FILES To use a summary file:
In a summary file, flows are 1. Choose Edit > Script > Start, and enter the following ksh command to specify where the GDE
represented by a kind value of flow. should create the summary file:
Their names consist of the name of export \
the component and the port to which AB_REPORT="$AB_REPORT summary=$RUN/$AB_GRAPH_NAME.summary.log"
the flow is attached. Special
where AB_GRAPH_NAME is an environment variable that the GDE automatically sets to
characters in the flow name are
replaced by underscores. represent the name of your graph, without the .mp extension, and with any special characters
(such as spaces) replaced by underscores.
2. Run the graph.
The tracking summary is written to the Audit File output component of the graph.
3. Right-click the Audit File and choose View Data to compare the contents of the summary file
with the original graph and find the flows you will need for your audit. In this example you
need to make sure that no records were lost between these flows:
Partition_by_Key_Account_Num.in
Join_Cust_Acct.out flows (one for partition 0 and one for partition 1)
Specifically, you need to verify that the sum of the num_records fields (specified in
summary.dml) for the Partition_by_Key_Account_Num.in flow is the same as the sum for the
Join_Cust_Acct.out flows. The ROLLUP used in this graph and described in the next section
accomplishes this.

114 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Using a ROLLUP to check the record counts
To use a rollup to check the record counts:
1. Create a new graph consisting of an INPUT FILE component, a ROLLUP component, and a TRASH
component, as shown in audit2.mp:

2. In the INPUT FILE component, specify the path to the previously generated summary file as the
URL:
file:$RUN/audit1.summary.log
3. Set the key on the ROLLUP to the empty key: { }
Here the output of the ROLLUP is unimportant. Its sole purpose is to force errors if the values
from the summary file do not match.
4. In the ROLLUP component, create variables to store the values from the summary file:
let decimal("") input_accounts = sum(in.num_records,
((in.kind=="flow") and
(in.name=="Partition_by_Key_Account_Num.in")));
let decimal("") output_accounts = sum(in.num_records,
((in.kind=="flow") and (in.name=="Join_Cust_Acct.out")));
Because the summary file uses a record format with conditional fields, in.name will exist only
for certain values of in.kind, so a condition such as in.kind=="flow" is required to avoid NULL

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 115

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 values. For information about conditional fields, see the topic Conditional fields (under DML
Reference > Compound Types > Records) in Ab Initio Help.
5. Check the input and output account variables for NULL values to catch unexpected errors:
if (is_null(input_accounts))
force_error("No records found in input file");
if (is_null(output_accounts))
force_error("No records found in output file");
Because the next statement tests for inequality, this is not strictly necessary (a NULL value
would fail anyway), but it provides a more informative error message.
6. Force an error and output an appropriate message if the counts of the input and output
accounts do not match:
if ((input_accounts!=output_accounts))
force_error(string_concat("Count of input accounts '",
input_accounts, "' does not match count of output accounts '",
output_accounts, "'"));
You can surround the values in the error string with single quotes for clarity.
7. Write out the values being compared. This typically makes transforms like this easier to debug
(and the ROLLUP will get errors if it does not assign an output):
out.input_accounts :: input_accounts;
out.output_accounts :: output_accounts;

Limitations of using a summary file and a ROLLUP

Using a summary file for auditing has the advantage that you do not complicate your original
graph with extra components to collect log information, but has the following limitations:
z A summary file can be used only to audit information written to the file, such as record
counts. The summary file does not contain financial totals, for example.

116 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 z Using a summary file requires the audit graph to run after the original graph, so
scheduling gets more complex and cluttered. Ideally, you would like the original graph
to fail if it fails the audit checks.
z Using a summary file creates a possible dependency between the two graphs: changes
to the first graph might necessitate changes to the auditing graph. In the best case,
failure to update the names in the second graph will cause it to fail after changes to the
first graph. The worst case is shown by the following graph (audit3.mp):

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 117

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

118 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Here an intentional bug has been added: the select_expr parameter in the FILTER BY EXPRESSION
component is set to zero, causing all the records to be dropped. But if you made this change and
did nothing else, the audit graph would report success, because it counts the output from the
JOIN. To see this, run this graph, change the input file in audit3.mp to point to file:$RUN/
audit3.summary.log, and then run audit3.mp. It still counts the output from the JOIN, so it
reports success even though the output file is empty. To make this graph do a proper audit, you
would need to change your out flow in audit3.mp to Filter_by_Expression.out, but that would
be easy to forget.

NOTE: If you use summary files for auditing, make sure the original graph includes a legend
documenting the need to change the audit graph if the original graph changes.

More complex auditing: Using log files

The more common, though more complex, method of auditing graphs is to add components to
the graph to collect the necessary data and audit the results as required.
This recipe describes how to create a self-contained graph (audit4.mp, shown below) that not
only produces output from its inputs, but also tests various assumptions about how the output
should relate to the input. It uses components that are completely separate from the processing
logic itself, and it handles odd conditions such as unexpected NULL values and the absence of
any input records.

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 119

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

120 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Specifically, this audit graph checks that: ANOTHER WAY TO COLLECT
z No input records have been dropped. DATA ABOUT INPUTS
A straightforward but less efficient
z Total balances are the same in the input and the output.
way to collect data about the inputs
z The total of all transaction amounts in the input equals the sum of the debits and would be to add a REPLICATE
credits in the output. component, make a second copy of
the input flow, and use a ROLLUP
To accomplish this, the graph does the following:
component to count the inputs
1. Uses log files to collect the relevant data from the inputs (page 121). and/or sum the financial data.
2. Rolls up the graphs output (page 124).
3. Reads back and rolls up the values from the log files (page 125).
4. Compares the input and output values (page 127).

NOTE: If you follow these steps to audit a similar graph, you should test the audit graph as
described in Testing the audit graph (page 129).

Collecting log data about the inputs

To collect log data about the inputs:
1. After partitioning each input file by key, add a REFORMAT component after each partitioner.
Adding the REFORMAT components after the partitioners (rather than attaching them directly
to the input files) is safe, as partitioners never reject records. This allows the collection of log
information to proceed in parallel.

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 121

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 2. Define a log_type as shown in the following code and save it in a DML file (for audit4.mp this
is log_type.dml) to be included later when you send information to the log ports of the four
REFORMATs:
type log_type =
record
string("") event_type;
string("") event_text;
end;
3. Do the following in the REFORMAT components:
a. For Customer File and Account File, create a transform that includes log_type.dml,
creates a global variable count to accumulate the count of customers and accounts, and
defines a function named final_log_output to send count to the log port, as shown in
the following code. (The function must be named final_log_output. For details, see
About the final_log_output function on page 124.) The graph audit4.mp does this in the
transform file audit_count.xfr:
include "/~$AI_DML/log_type.dml";

let decimal("") count =0;

log_type out::final_log_output() =
begin
out.event_type :: "audit";
out.event_text :: count;
end;

out::reformat(in) =
begin
count = count + 1;
out.* :: in.*;
end;

122 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. For the Balance file, you need to accumulate the count and the sum, so write an
embedded transform that includes log_type.dml, creates global variables count and
sum_balances, and defines a function named final_log_output to send the values to the
log port, as shown in the following code. (Again, the function must be named
final_log_output. For details, see About the final_log_output function on page 124.)
include "/~$AI_DML/log_type.dml";

let decimal("") count =0;

let decimal("") sum_balances =0;

log_type out::final_log_output() =
begin
out.event_type :: "audit";
out.event_text :: string_concat(count, "|", sum_balances);
end;

out::reformat(in) =
begin
count = count + 1;
sum_balances = sum_balances + first_defined(in.bal, 0);
out.* :: in.*;
end;
c. For the Transaction file, you also need to accumulate the count and the sum, so write
an embedded transform that includes log_type.dml, creates global variables count and
sum_tran, and defines a function named final_log_output to send them to the log port,
as shown in the following code. (Again, the function must be named final_log_output.
For details, see About the final_log_output function on page 124).
include "/~$AI_DML/log_type.dml";

let decimal("") count =0;

let decimal("") sum_tran =0;

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 123

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

ABOUT THE final_log_output log_type out::final_log_output() =
begin
FUNCTION
out.event_type :: "audit";
In any component with a transform, if
out.event_text :: string_concat(count, "|", sum_tran);
a function named final_log_output is end;
present, it is called after the transform
has been called for all input records, out::reformat(in) =
and it writes its result to the log port, begin
using any values accumulated in count = count + 1;
global variables. (For details on this sum_tran = sum_tran + first_defined(in.amt, 0);
function, search Ab Initio Help for out.* :: in.*;
final_log_output function.) end;

Rolling up the graphs output

To perform a meaningful audit, you must check the inputs recorded in your log data against the
appropriate values in the graphs outputs. To accomplish this, you need to roll up the outputs to
a single count and (for balances and transactions) a single total, as demonstrated in the Audit
Totals subgraph in audit4.mp and described in this section. These ROLLUPs yield the needed data
as follows:
z They count and sum the non-NULL balances in the output, which should equal the
counts and sums of the input balance records.
z Since each transaction contributed to either a credit count and sum or a debit count
and sum, the rolled-up counts and sums of the debits and credits on the output should
equal the transaction counts and sums on the input.
To roll up the graphs output:
1. Roll up the graphs output on cust_id to find distinct customers.

124 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

2. Roll up the output of the first ROLLUP in parallel on the empty key: { } to generate a single
grand total for each input.
Doing these two ROLLUPs is more efficient than rolling up directly from the account level,
because many customers have multiple accounts, so there are fewer records after the ROLLUP
to the customer level. Also, because the input to the first ROLLUP is at the account level and
the second is at the customer level, and both inputs might be a lot of records, these ROLLUPs
should be done in parallel.
3. Roll up the output of the second ROLLUP in serial to generate a single grand total from the
totals of the individual partitions.
Since the data is already partitioned by cust_id, there is no double-counting. Note also that
layouts of these last two ROLLUPs are set explicitly to prevent propagation from causing
something to run in the wrong layout.

Reading back and rolling up the values from the log files
Before your subgraph can audit the results by comparing the log values with the output values
from the main graph, you must generate the needed log values (those with an event_type of
audit) in the appropriate format. Also, since the log files were attached to multiple components
executing in parallel, you need to roll these up to get the total values for comparison.
To do this, write separate transforms to read back and roll up the log values for the following
two cases:
z Customer and Account logs count values
z Balance and Transaction logs count and sum values

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 125

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Customer and account logs (count values) Do this ROLLUP by putting the code shown below in your
transform. In audit4.mp, this code is in sum_log_counts.xfr. Note that because the event_text
field in log files is stored as a string delimited by |\n, you must cast it to a decimal so that you
can use the data as a number.
out.sum_counts :: sum((decimal(""))in.event_text,
in.event_type == "audit");

Balance and transaction logs (count and sum values) Reading back the log values for the balances
and transactions is more complex, because two values separated by "|" are stored in a single
in.event_text field in the log file. You can use the reinterpret_as function to interpret this string
as a record containing those two values. In audit4.mp, the code for this is in
sum_log_counts_values.xfr.
To read back the log values for the balances and transactions:
z Create an expression with a record format that matches the format used for the event_text in
the log file, and assign this expression to a variable as follows:
a. On the Variables tab of the ROLLUP transform, add the following expression:
reinterpret_as(record decimal("|") count; decimal("|\n") sum;
end, in.event_text);
b. Drag the expression across to the Variables output pane.
The GDE creates a variable of type record with a default name such as expr1 and the
following format, and assigns the expression to it:
record
decimal("|") count;
decimal("|\n") sum;
end

126 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 c. Rename the variable appropriately (log, for example) by clicking it twice (not double- MAKING YOUR GRAPHS AND
clicking) and typing a new name. CODE MORE READABLE
In Text View, the resulting code looks like this: You can change the default names
out::rollup(in) = assigned to subgraph ports and
begin numbered ports as follows:
let record 1. On the Ports tab of the
decimal("|") count; component or subgraph, right-
decimal("|\n") sum; click the name of the port.
end log = 2. Click Rename.
reinterpret_as(record
3. Type a name for the port and
decimal("|") count;
press Enter.
decimal("|\n") sum;
end,
in.event_text);
d. Roll up log.count and log.sum and finish the transform as follows:
out.sum_counts :: sum(log.count, in.event_type == "audit");
out.sum_values :: sum(log.sum, in.event_type == "audit");
end;

Comparing the input and output values

After rolling up the graphs output data and the generated log data, you have five flows, each
rolled up to a single record containing totals: one for the graphs output and four for the log
outputs.
To check the values (as shown in the Audit Totals subgraph in audit4.mp):
1. Do an explicit join of the five flows.

TIP: Change the names of the in ports on the JOIN component to make them more
descriptive.

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 127

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 2. To ensure that the JOIN transform gets called (has something to join against) if some or all of
these flows have no records, so that you can flag NULL flows as errors, send the log port of
the second ROLLUP (parallel rollup to grand total) into the JOIN, and give the port an
appropriate name. In the Audit Totals subgraph, this is the rollup_log port.

NOTE: The log port of a transform always has one or more records, even if the transform did
not receive any input records.

TIP: Another way of ensuring that a transform gets called is to add a GENERATE RECORDS
component to create a single dummy record.
3. Set the record_required parameter of the log ports flow into the JOIN to True (record
required on this input). (In the Audit Totals subgraph, this is record_required1.)
4. To ensure that the log ports flow has exactly one record, set the appropriate dedup
parameter of the JOIN to Dedup this input before joining. (In the Audit Totals subgraph, this
is dedup1.)
5. Test for NULL and force an error whenever the counts and totals on the inputs and outputs do
not match as expected.
In other words, write a sequence of conditions of the form if (error condition)
force_error("error message"). For example:
if (is_null(totals))
force_error("No records in output dataset");
...
if ((totals.debits + totals.credits) != transactions.sum_values)
force_error(string_concat(
"Sum of input transaction amounts '",
(decimal("")) transactions.sum_values,
"' does not match sum of output debits and credits '",
(decimal("")) (totals.debits + totals.credits),
"'"
));

128 Auditing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

6. To aid in debugging (and to provide outputs, which are required for every JOIN), assign the
counts and sums from the grand total to the outputs:
out.* :: totals.*;

Testing the audit graph

Graphs that involve this sort of auditing require extensive testing. It is not sufficient just to run
some legitimate test data. You should also create test conditions to trigger every possible error
message (even if you think some conditions cannot possibly occur) and make sure that the
correct error is actually detected in each case. This may be impossible to do using just the
original graph, since you coded it to handle all errors.
Often a useful technique is to add a component directly before your output file to cause various
errors. For example, you could add a FILTER BY EXPRESSION to drop one (or all) records, or a
REFORMAT to change the value of a single field:
out.foo :: if (next_in_sequence() == 500) 0 else in.foo;

AB INITIO CONFIDENTIAL AND PROPRIETARY Auditing 129

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Processing sets of files
In many situations, an input to a graph is not a single file, but several files. The files may hold
data collected from different locations (such as daily cash register data from each of several
stores), at different times (hourly logs of a Web server), or both.
The recipes in this chapter present two approaches to processing such data:
z Using ad hoc multifiles to process multiple files in data parallel fashion
z Using READ MULTIPLE FILES and WRITE MULTIPLE FILES components to read and write
multiple files in a single processing stream
The two simplest approaches to processing sets of files both have disadvantages in certain
situations:
z Running the same graph multiple times, once for each file
If the files can be processed independently, this may be acceptable if there are not very
many of them. If they are numerous, however, the overhead of starting up the graph
once for every file can be excessive.
z Building a graph that processes a fixed number of files, with one INPUT FILE component
for each
This approach (shown in hardwired.mp below) may allow some parallelism, but it is
inherently inflexible, as the graph can process only the number of files specified when
the graph was developed.

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 131

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Ad hoc multifiles
This section presents four graphs (provided in the file-sets sandbox) that show they all produce
the same output. But each graph demonstrates a different way of specifying multiple files, as
described in the following subsections:
z Listing files explicitly (ad_hoc_explicit.mp, page 134)
z Using wildcards to list files (ad_hoc_wild_card.mp, page 135)
z Using a command to list files (ad_hoc_command.mp, page 137)
z Using a parameter to list files (ad_hoc_parameter.mp, page 137)

132 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 NOTE: One potential downside to ad hoc multifiles is that if a very large number of partitions
(hundreds or thousands) are being processed, the graph may end up running hundreds or
thousands of ways parallel. This could stress or exhaust system resources.
Each of the following graphs treats three serial input files as a multifile, reformats the data in the
description field into a one-character trans-kind field, interleaves the partitions, and sends the
result to a single serial file. For example:

All three input files have the same record format, which is based on the user-defined type
specified in transaction_type.dml:
type transaction =
record
date("YYYY.MM.DD") trans_date;
decimal(9,0) account_id;
decimal(10.2) amount;
string('\n') description;
end;

The input files use this by including transaction_type.dml in the record format and identifying
transaction as the top-level type:
include "/~$AI_DML/transaction_type.dml";
metadata type = transaction;

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 133

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The input data consists of three files containing transaction records for three months:
z trans-2006-01.dat
z trans-2006-02.dat
z trans-2006-03.dat

Listing files explicitly

To define an ad hoc multifile by listing files explicitly:
1. On the Description tab of the Properties dialog for the INPUT FILE, click Edit.
2. In the Define Multifile Partitions dialog, click New and enter the full path of the first file you
want to process. For example:
$AI_SERIAL/trans-2006-01.dat
3. Repeat Step 2 analogously for each additional file, and click OK.
In ad_hoc_explicit.mp, the Define Multifile Partitions dialog looks like this:

134 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

4. Click OK twice to close the Define Multifile Partitions dialog and the Properties dialog.
When you run the graph, the INPUT FILE behaves like a three-way multifile whose data partitions
are the three listed files; components can run in the layout of the INPUT FILE component.

Using wildcards to list files

Listing files by using wildcards is one of the most common ways to specify an ad hoc multifile,
because often the names of the files to be processed are similar but are not precisely known
until the graph is run.

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 135

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 To define an ad hoc multifile by using wildcards:
1. On the Description tab of the Properties dialog for the INPUT FILE, click Edit.
2. In the Define Multifile Partitions dialog, click New and enter the full path of the files you
want to process, using an asterisk (*) as a wildcard to represent any sequence of characters
that is not identical in the names of all files you want to process. For example:
$AI_SERIAL/trans*.dat
In ad_hoc_wild_card.mp, the Define Multifile Partitions dialog looks like this:

3. Click OK twice to close the Define Multifile Partitions dialog and the Properties dialog.

136 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Using a command to list files
A shell command is useful if you want to:
z Generate an ad hoc multifile based on file properties other than their names
z Get a list of files from another program
The graph ad_hoc_command.mp shows the simple case of using the ls command to generate an
ad hoc multifile. This approach is not recommended (using a wildcard as described on page 135
would be easier), but it demonstrates the technique.
To define an ad hoc multifile by using a command:
1. On the Description tab of the Properties dialog for the INPUT FILE, click Edit.
2. In the Define Multifile Partitions dialog, click New and enter the shell command that will
generate a list of the files you want to process. For example:
$(ls $AI_SERIAL/ad_hoc_input_*.dat)

NOTE: You must enclose the command in parentheses and precede the opening parenthesis
with a dollar sign: $(...) to indicate that what follows is a ksh command.
3. Click OK twice to close the Define Multifile Partitions dialog and the Properties dialog.

Using a parameter to list files

You can also define an ad hoc multifile by using a graph parameter. This is useful, for example, if
an external process is being used to provide a list of the input files.
To define an ad hoc multifile by using a parameter:
1. Choose Edit > Parameters.
2. In the Edit Parameters dialog, choose Edit > Insert.

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 137

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 3. Add a graph parameter (for example, LIST_OF_FILES) and make sure that its interpretation is
set to $ substitution.
4. Click the pencil icon to the right of the parameter name.
5. In the Editing name parameter dialog, type the full paths of the input files, separated by
spaces or carriage returns, and save the parameter.
In ad_hoc_parameter.mp, the LIST_OF_FILES parameter is defined as follows:

6. Click OK to close the Edit Parameters dialog.

Ad hoc output multifiles

Ad hoc output multifiles are useful when a downstream processing system needs to receive a set
of files rather than a single serial file or a multifile. The graph ad_hoc_explicit_output.mp
demonstrates this.

138 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 NOTE: Because both the input and output are partitioned three ways, the REFORMATs
processing will run three-ways parallel one way for each partition.
To use an explicitly defined ad hoc output multifile, follow the same steps you would for an
input file (see page 134).

READ MULTIPLE FILES and WRITE MULTIPLE FILES

components
The recipes in this section demonstrate and describe the following techniques:
z Reading multiple files (below) read_multiple_files_simple.mp reads three input files,
reformats them to use a different record type, and writes the file contents to a single
serial file.
z Reading and writing multiple files (page 144) read_and_write_multiple_files.mp
reads three input files, reformats them to use a different record type and add a field
containing the filename, and writes the file contents to three appropriately named
serial files.

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 139

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Reading multiple files
The graph read_multiple_files_simple.mp reads the files specified in a list and writes their
contents to a single serial file:

To read multiple files from a list and write their contents to a single serial file:
1. Add an INPUT FILE component whose URL specifies a file containing a list of the files to read
and whose record format specifies how the filenames are separated.
In read_multiple_files_simple.mp, the URL of Input File List is file:$AI_SERIAL/
list_of_files.dat, and the referenced file has the following content:
trans-2006-01.dat
trans-2006-02.dat
trans-2006-03.dat
The record format for list_of_files.dat files is:
record
string('\n') filename;
end
2. Connect a READ MULTIPLE FILES component to the INPUT FILE and configure it as follows:
a. Hold down the Shift key and double-click the component to generate a default
transform. (If the Transform Editor is displayed in Package View, switch to Text View,
and you will be prompted to generate the transform.)

140 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. Delete all the transform code except the get_filename function. USER-DEFINED TYPES
c. Code the get_filename function such that it will generate the full path to each input file Creating a user-defined type such as
transaction in a separate DML file
and then retrieve its content. For example:
allows you to reuse the type in
filename :: get_filename(in) =
different ways. You can use it as the
begin
top-level type in a record format, as
filename :: string_concat($AI_SERIAL, '/', in.filename);
in transaction.dml, or include it in a
end;
transform function, as described in
d. On the Ports tab of the Properties dialog for the READ MULTIPLE FILES component, Reading and writing multiple files
specify the record format for the output data (which is the record format for each file in (page 144).
the list). To do this, select Use file and then browse to the location of a file containing
the record format.
This graph uses transaction.dml, which includes transaction_type.dml. The definition
of transaction in transaction_type.dml is as follows:
type transaction =
record
date("YYYY.MM.DD") trans_date;
decimal(9,0) account_id;
decimal(10.2) amount;
string('\n') description;
end;
The file transaction.dml declares the named type transaction as a top-level type for
use as the output record format:
include "/~$AI_DML/transaction_type.dml";
metadata type = transaction;
The graph read_and_write_multiple_files.mp, described on page 144, also uses this
type: its READ MULTIPLE FILES component uses it as the input_type and as part of the
output record format.

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 141

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 3. Attach a REFORMAT component to the READ MULTIPLE FILES component and code its transform
appropriately. To code it like the REFORMAT in read_multiple_files_simple.mp, do the
following:
a. Include the transform file map_trans_kind.xfr, which replaces the data in each
description field with a single character as follows:
out :: map_trans_kind(description) =
begin
out :1: if (string_index(description, "ATM")) "A";
out :2: if (string_index(description, "Teller")) "T";
out :3: if (string_index(description, "Check")) "C";
out :: "O";
end;
b. Output all fields, mapping the reformatted description field to a trans_kind field in the
output. The complete REFORMAT transform in read_multiple_files_simple.mp looks like
this:
include "/~$AI_XFR/map_trans_kind.xfr";

out::reformat(in) =
begin
out.* :: in.*;
out.trans_kind :: map_trans_kind(in.description);
end;
4. Attach an OUTPUT FILE component to the REFORMAT component and configure it as follows:
a. On the Description tab of the Properties dialog for the OUTPUT FILE, specify the URL for
the serial file to which you want to write the output data. For example:
file:$AI_SERIAL/processed.dat
b. On the Ports tab of the Properties dialog for the OUTPUT FILE, specify the record format
for the output data by selecting Use file and then browsing to the location of a file
containing the record format.

142 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 This graph uses processed.dml, which includes processed_type.dml. The definition of
processed in processed_type.dml is as follows:
type processed =
record
date("MMM DD, YYYY") trans_date;
decimal(9,0) account_id;
decimal(10.2) amount;
string('\n') trans_kind;
end;
The file processed.dml declares the named type processed as a top-level type for use as
the output record format:
include "/~$AI_DML/processed_type.dml";
metadata type = processed;
Again, read_and_write_multiple_files.mp, described on page 144, uses this type in a different
way: its REFORMAT component uses the type as part of an embedded output format that also
includes the filename for each record, and its WRITE MULTIPLE FILES component uses it as the
output_type.
5. Run the graph and view the results.
The output should look like this:

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 143

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Reading and writing multiple files
The graph read_and_write_multiple_files.mp reads the contents of the same three input files as
read_multiple_files_simple.mp (described on page 140). However, rather than simply writing
the reformatted records to a single output file, it uses READ MULTIPLE FILES and WRITE MULTIPLE
FILES components in combination to get the filenames and write the records to three
appropriately renamed files. It shows how to:

144 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 z Include data from the input to a READ MULTIPLE FILES component in its output records. In
this example, the filenames are added to each output record, and the output format is
different from the record format of the files that the component reads.
z Write multiple files, using a different record format than that of the data flowing into
the WRITE MULTIPLE FILES component. In this example, the output records of the
REFORMAT include the filename, but the records written by the WRITE MULTIPLE FILES
component do not. Instead, the component uses that information to name the files it
writes to.

To read multiple files from a list and write reformatted data to renamed output files:
1. Add an INPUT FILE component whose URL specifies a file containing a list of the files to read
and whose record format specifies how the filenames are separated.
In read_and_write_multiple_files.mp, the URL of Input File List is file:$AI_SERIAL/
list_of_files.dat, and the referenced file has the following content:
trans-2006-01.dat
trans-2006-02.dat
trans-2006-03.dat

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 145

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The record format for list_of_files.dat is:
record
string('\n') filename;
end
2. Connect a READ MULTIPLE FILES component to the INPUT FILE and configure it as follows:
a. Hold down the Shift key and double-click the READ MULTIPLE FILES component to
generate a default transform. (If the Transform Editor is displayed in Package View,
switch to Text View, and you will be prompted to generate the transform.)
b. Specify the record format for reading the records from the input files:
include "/~$AI_DML/transaction_type.dml";
type input_type = transaction;
c. Code the get_filename function such that it will generate the full path to each input file
and then retrieve its content. For example:
filename :: get_filename(in) =
begin
filename :: string_concat($AI_SERIAL, '/', in.filename);
end;
d. Code the reformat function to output the name of each input file and all desired fields.
This graph outputs the fields defined by transaction in the output record format as
described in Step e:
out::reformat(in, filename) =
begin
out.filename :: filename;
out.trans :: in;
end;

146 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 e. On the Ports tab, select Embed and specify the record format for the output.
This graph includes the file transaction_type.dml, which contains the type definition of
transaction (page 141), and creates a metadata type consisting of fields for it and the
filename:
include "/~$AI_DML/transaction_type.dml";
metadata type =
record
string(",") filename;
transaction trans;
end;
3. Connect a REFORMAT to the READ MULTIPLE FILES component and configure it as follows:
a. On the Ports tab, select Embed and specify the record format for the output.
This graph includes the file processed_type.dml, which contains the type definition of
processed (page 143), and creates a metadata type consisting of fields for it and the
filename:
include "/~$AI_DML/processed_type.dml";
metadata type =
record
string(",") filename;
processed trans;
end;
b. Code the transform as follows to output the filenames and the processed fields.
This graph replaces the description field of the transaction type with the trans_kind
field of the processed type as follows:
include "/~$AI_XFR/map_trans_kind.xfr";

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 147

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 out::reformat(in) =
begin
out.filename :: in.filename;
out.trans.* :: in.trans.*;
out.trans.trans_kind :: map_trans_kind(in.trans.description);
end;
4. Connect a WRITE MULTIPLE FILES to the REFORMAT component and code its transform as follows:
a. Specify the record format of the output type.
This graph includes the DML file containing the definition of the desired output type,
processed:
include "/~$AI_DML/processed_type.dml";
type output_type = processed;
b. Name the output files appropriately:
This graph replaces the string trans from the input filenames with processed in the
output filenames:
filename::get_filename(in) =
begin
filename :: string_replace(in.filename, 'trans',
'processed');
end;
c. Write the reformatted transaction data to each output file:
write::reformat(in) =
begin
write :: in.trans;
end;
The WRITE MULTIPLE FILES component uses the result of the reformat function as the data
to write to the file.

148 Processing sets of files AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

5. If you want to be able to view the contents of one or more of the input files listed in Input File
List, add an INPUT FILE component and configure it as follows:
a. Set its URL to the location of the input file.
In read_and_write_multiple_files.mp, this is file:$AI_SERIAL/trans-2006-01.dat.
b. Disable it to prevent others from thinking it is an actual input.
6. If you want to be able to view the contents of one or more of the output files written by the
WRITE MULTIPLE FILES component, add an OUTPUT FILE component and configure it as follows:

a. Set its URL to the location of the output file.

In read_and_write_multiple_files.mp, this is file:$AI_SERIAL/processed-2006-01.dat.
b. Disable it to prevent others from thinking it is an actual output.
In read_and_write_multiple_files.mp, the data for the first output file looks like this:

AB INITIO CONFIDENTIAL AND PROPRIETARY Processing sets of files 149

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Header and trailer record processing
While most datasets contain only one kind of record, some (particularly those from legacy
systems) may contain several. The most common form of such data consists of groups of
records that each have a header record, followed by some number of body records, followed by
a trailer record. The header record usually contains information that applies to each of the body
records, and the trailer often contains summary information. For example, a dataset could
consist of groups of transactions, with each group made up of a header record containing
customer information, body records for each transaction, and a trailer record containing totals.
This chapter begins by describing three different formats for representing data that consists of
headers, bodies, and trailers. It then describes graphs that process data in those formats:
z Generating separate summaries of header and body subrecords (page 160)
z Distributing header information to body records (page 165)
z Normalizing vector data to header and body records (page 167)
z Validating flat input data (page 169)
z Validating vector input data (page 174)
Finally, an advanced topic (page 176) describes a graph that constructs header/body/trailer data
from simple records.

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 151

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Input data formats
With Ab Initio DML, the same data can often be described in many ways. The graph ht_data.mp
(in the header sandbox) shows three ways of describing the same physical data:
z Flat record format (page 153)
z Subrecord format (page 155)
z Vector format (page 158)
Rather than running ht_data.mp (it does not do anything), you should view the record format
and the data for each INPUT FILE component in it to see the differences in how the records are
described and displayed.

152 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Flat records
The records in the dataset ht1_input Flat Records (in ht_data.mp) are represented in a flat
format (defined in ht1_input.dml) that uses conditional fields. The presence or absence of a
value in each conditional field is determined by the value of the kind field:
record
string(1) kind; /* One of H, B or T */
if (kind=="H")
begin
decimal(',') transaction_id;
string(',') customer_name;
decimal('\n') num_items;
end;
if (kind=="B")
begin
string(',') item;
decimal(',') quantity;
decimal('\n') cost;
end;
if (kind=="T")
begin
decimal(',') total_quantity;
decimal('\n') total_cost;
end;
end

For each customer transaction, this format results in at least two records (one header and one
trailer) with all nine fields in each record, and NULL values for each empty field. To see this, view
the data in ht1_input Flat Records:

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 153

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

154 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Subrecords
The records in the dataset ht1_input Using Subrecords (in ht_data.mp) are represented in a
format that uses subrecords as follows:
z ht1_input_subrec_types.dml defines three types: header_type, body_type, and
trailer_type
z ht1_input_subrec.dml includes ht1_input_subrec_types.dml and defines a record
format with three fields, one of each of the three types defined in
ht1_input_subrec_types.dml
ht1_input_subrec_types.dml
type header_type =
record
decimal(',') transaction_id;
string(',') customer_name;
decimal('\n') num_items;
end;

type body_type =
record
string(',') item;
decimal(',') quantity;
decimal('\n') cost;
end;

type trailer_type =
record
decimal(',') total_quantity;
decimal('\n') total_cost;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 155

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 ht1_input_subrec.dml
include "/~$AI_DML/ht1_input_subrec_types.dml";
metadata type =
record
string(1) kind; /* One of H, B or T */
if (kind=="H") header_type header;
if (kind=="B") body_type body;
if (kind=="T") trailer_type trailer;
end;

For each customer transaction, this format also results in at least two records (one header and
one trailer) with three subrecords in each record, and NULL values for each subrecord that has
no content. To see this, view the data in ht1_input Using Subrecords:

156 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 157

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Vectors
In some cases, a header/body/trailer group can be treated as a single record, with a vector used
to hold the body (sub)records.
The records in ht1_input Using Vectors (in ht_data.mp) are represented in a vector format, in
which the number of body records is determined by the value of the num_items field in the
header record:
ht1_input_vector.dml
record
record
string(1) kind; /* Should be H */
decimal(',') transaction_id;
string(',') customer_name;
decimal('\n') num_items;
end header;
record
string(1) kind; /* Should be B */
string(',') item;
decimal(',') quantity;
decimal('\n') cost;
end body[header.num_items];
record
string(1) kind; /* Should be T */
decimal(',') total_quantity;
decimal('\n') total_cost;
end trailer;
end

As you can see by viewing the data in ht1_input Using Vectors, this format results in just one
record for each customer transaction: it consists of the header data, a vector of body data, and
one trailer record, and it has no NULL fields:

158 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 159

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Processing the data
This section describes ways of generating summaries of data stored in subrecord or vector
format:
z Generating separate summaries of header and body subrecords (next)
z Distributing header information to body records (page 165)
z Normalizing vector data to header and body records (page 167)

Generating separate summaries of header and body subrecords

The graph ht_process1.mp processes the subrecord view of the data to generate summaries of
the customer data and the product data:

160 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

The graph does this as follows:
1. Uses a REFORMAT to split out the different subrecord types:
a. Sets the output_index parameter as follows so that header records (those with kind H)
will be sent to the first output's transform function (transform0), body records to the
second (transform1), and trailer records to the third (transform2):
out :: output_index(in) =
begin
out :: if (in.kind == 'H') 0 else if (in.kind == 'B') 1
else 2;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 161

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. Defines the following three transforms to produce the respective outputs:
transform0 transform1 transform2
out :: reformat(in) = out :: reformat(in) = out :: reformat(in) =
begin begin begin
out :: in.header; out :: in.body; out :: in.trailer;
end; end; end;

For example, transform0 looks like this in Grid View:

162 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

2. Uses two ROLLUP components to summarize the following data in the output:
Customer data (number of store visits and number of items bought by each customer):
out::rollup(in) =
begin
out.customer_name :: in.customer_name;
out.num_store_visits :: count(1);
out.total_num_items :: sum(in.num_items);
end;

Product data (item, quantity sold, and average cost):

out::rollup(in) =
begin
out.item :: in.item;
out.total_quantity :: sum(in.quantity);
out.average_cost :: avg(in.cost);
end;
3. Sends the trailer records to a TRASH component.

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 163

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The Customer Report output of ht_process1.mp looks like this:

164 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Distributing header information to body records
The graph ht_process2.mp uses a SCAN to process the subrecord view of the data and generate a
dataset containing records of a single type that combines information from header and body
records.

The SCAN does this as follows:

1. Sets the key-method parameter to Use key_change function and defines the function as
shown below. This causes the SCAN to execute its initialize function whenever it encounters a
record whose kind field contains H.
out :: key_change(previous, current) =
begin
out :: current.kind == 'H';
end;
2. Defines the temporary_type as a record consisting of fields for the transaction ID and
customer name:
type temporary_type =
record
decimal('') transaction_id;
string('') customer_name;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 165

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The temporary variable is used in the scan function (see Step a in Step 4 below) to carry
information from the header record to each of the other records in the group.
3. For each input record whose kind is H (and which therefore denotes the start of a new key
group), the SCAN initializes the temporary record to hold the contents of the two fields in the
header record that match the fields in the temporary_type (that is, transaction_id and
customer_name):
temp :: initialize(in) =
begin
temp :: in.header;
end;
4. For all records, the SCAN does the following:
a. Returns a temporary record with the content of the current temporary record:
out :: scan(temp, in) =
begin
out :: temp;
end;
This causes the current transaction_id and customer_name fields from the header
record to be written to the temporary record for each subsequent body and trailer
record in the key group initialized for that header record.
b. Executes its finalize function, outputting the content of the temporary variable and the
values of the item, quantity, and cost fields:
out :: finalize(temp, in) =
begin
out.transaction_id :: temp.transaction_id;
out.customer_name :: temp.customer_name;
out.item :: in.body.item;
out.quantity :: in.body.quantity;
out.cost :: in.body.cost;
end;

166 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 NOTE: The SCANs reject-threshold parameter is set to Never abort to prevent the graph from
failing when the SCAN tries to assign the in.body fields item, quantity, and cost (which do not
exist in the header or trailer records) to the correspondingly named output fields.
The output of ht_process2.mp looks like this:

Normalizing vector data to header and body records

The graph ht_process3.mp (in the headers sandbox) produces the same output as
ht_process2.mp (a dataset containing records of a single type that combines information from
header and body records) but uses a NORMALIZE to process the vector view of the data:

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 167

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The NORMALIZE does the following:
1. Uses the num_items field from each header record to specify the number of body records to
output for each transaction:
out :: length(in) =
begin
out :: in.header.num_items;
end;
2. Outputs (for each transaction) a record consisting of the header fields transaction_id and
customer_name, along with the body fields item, quantity, and cost.
out :: normalize(in, count) =
begin
out.transaction_id :: in.header.transaction_id;
out.customer_name :: in.header.customer_name;
out.item :: in.body[count].item;
out.quantity :: in.body[count].quantity;
out.cost :: in.body[count].cost;
end;

168 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Validating the input data
In some situations you may need to assess the quality of data across a group of records
consisting of a header record, some number of body records, and a trailer record. Exactly how to
do this depends on the representation of the data. The following sections describe how to assess
the quality of data represented in two ways:
z Validating flat input data (next)
z Validating vector input data (page 174)

Validating flat input data

The graph ht_validate1.mp shows how to validate the flat representation of the input data
described on page 153:

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 169

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

NULL VALUES IN ROLLUP The components of ht_validate1.mp do the following:
AGGREGATION FUNCTIONS 1. The Rollup: Validate and Summarize component does what its name suggests, as follows:
In the conditional representation used
for the flat data (see Flat records a. Sets the key-method parameter to Use key_change function, and defines the function
on page 153), fields that are not part as shown below. This causes the ROLLUP to treat each set of header, body, and trailer
of a given kind of record are NULL. records (which begins with a record whose kind field is H) as a separate group.
(For example, the quantity field of all out :: key_change(previous, current) =
header and trailer records in ht_input begin
Flat Records is NULL.) Aggregation out :: current.kind == 'H';
functions such as sum handle these end;
records by simply ignoring NULL b. Creates variables to hold (for each group of header, body, and trailer records) the totals
values. calculated for the different items purchased, the quantity, and the cost:
first_defined FUNCTION out :: rollup(in) =
begin
The first_defined function returns
let decimal('') actual_num_items =
the first non-NULL argument
first_defined(count(in.item), 0);
passed to it. Here, it ensures that let decimal('') actual_total_quantity =
for any groups that do not have any first_defined(sum(in.quantity), 0);
body records, zero will be returned let decimal('') actual_total_cost =
instead of NULL. first_defined(sum(in.quantity * in.cost), 0);

first AND last AGGREGATION c. Outputs fields from the header and trailer records, along with booleans indicating
FUNCTIONS whether any of the calculated values in the original data are incorrect:
The first and last functions return out.transaction_id :: first(in.transaction_id);
values from the first and last records out.customer_name :: first(in.customer_name);
in a key group. In ht_validate1.mp, out.mismatched_num_items :: actual_num_items !=
these are the header and trailer first(in.num_items);
out.mismatched_total_quantity :: actual_total_quantity !=
records, respectively.
last(in.total_quantity);
out.mismatched_total_cost :: actual_total_cost !=
last(in.total_cost);
out.num_items :: first(in.num_items);
out.total_quantity :: last(in.total_quantity);

170 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 out.total_cost :: last(in.total_cost);
end;
2. The FILTER BY EXPRESSION component sets its select_expr parameter as follows, so that its out
port emits only records for which at least one of the mismatched fields is true:
mismatched_num_items or
mismatched_total_quantity or
mismatched_total_cost
Records that do not have mismatched numbers are sent to the FILTER BY EXPRESSIONs deselect
port, and from there to the Transaction Summary output file:

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 171

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 3. The REPLICATE copies the output of the FILTER BY EXPRESSION so that it can be used for two
purposes:
To generate a report containing the unmodified output of the FILTER BY EXPRESSION
that is, all transactions with mismatched totals. (See the Bad Transactions output
dataset below.)
To calculate the total number of records for each type of mismatch (see Step 4).

172 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

4. The Rollup: Build Quality Report component calculates the sums of the mismatched fields
from all records and sends them to the Quality Report output dataset:
out::rollup(in) =
begin
out.mismatched_num_items_count ::
sum(in.mismatched_num_items);
out.mismatched_total_quantity_count ::
sum(in.mismatched_total_quantity);
out.mismatched_total_cost_count ::
sum(in.mismatched_total_cost);
end;
The output looks like this:

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 173

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Validating vector input data
The graph ht_validate2.mp shows how to validate the vector representation of the input data
described on page 158:

This graph is identical to ht_validate1.mp (page 169) except that instead of using a ROLLUP
(which processes groups of records), this graph uses a REFORMAT, because each group of header,
body, trailer is in a single record.
The transform in the Reformat: Validate and Summarize component does the following:

174 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

1. Creates variables to hold the correct values for the total quantity of each item purchased and
the total cost for each store visit (that is, actual values that the graph will compute) and
initializes them to zero:
out :: reformat(in) =
begin
let decimal('') actual_total_quantity = 0;
let decimal('') actual_total_cost = 0;
2. Uses a for loop to iterate over the elements of each body vector and accumulate
actual_total_quantity and actual_total_cost:
let int i;

for (i, i < length_of(in.body))

begin
actual_total_quantity = actual_total_quantity +
in.body[i].quantity;
actual_total_cost = actual_total_cost + (in.body[i].quantity *
in.body[i].cost);
end
3. Outputs fields from the header and trailer records, along with boolean values indicating
whether any of the calculated values in the original data are incorrect:
out.transaction_id :: in.header.transaction_id;
out.customer_name :: in.header.customer_name;
out.mismatched_total_quantity :: actual_total_quantity !=
in.trailer.total_quantity;
out.mismatched_total_cost :: actual_total_cost !=
in.trailer.total_cost;
out.num_items :: in.header.num_items;
out.total_quantity :: in.trailer.total_quantity;
out.total_cost :: in.trailer.total_cost;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 175

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Advanced topic: Building flat data into header/body/trailer
data subrecord format
In some situations you may need to transform data from a homogeneous form into a header,
body, trailer form for example, when the results of a graph are to be fed back to a legacy
system. This section describes ht_build.mp, a graph that converts data from flat format into
subrecord format:

The record format of the input data is defined by ht_items.dml as follows:

record
decimal(',') transaction_id;
string(',') customer_name;
string(',') item;
decimal(',') quantity;
decimal('\n') cost;
end

176 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

The preceding format does not include the following fields from the headers and trailers of the
graphs described earlier in this chapter:
decimal(',') num_items; /* from the header */
decimal(',') total_quantity; /* from the trailer */
decimal('\n') total_cost; /* from the trailer */

The graph ht_build.mp computes those values as part of its processing. Specifically, it does the
following:
1. Rolls up the input records to compute the data for the above three fields (page 178).
2. Reformats the data to generic records that contain all eight fields, plus a kind field with a
value of H, B, or T to designate each record as a header, body, or trailer; and an ordering field
with a value of 1, 2, or 3 (page 179). The graph does this in two steps:
a. Reformats the body records
b. Reformats the rolled-up header records and trailer records
3. Merges the reformatted records, assembling them in header, body, trailer order (page 183).
4. Reformats the merged records, using a conditional record format to populate the header,
body, and trailer fields appropriately in the output data (page 184).

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 177

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Rolling up to compute header and trailer data
The Rollup Computing Header and Trailer Data component calculates (for each group of records
with the same {transaction_id}) the number of different types of items purchased, the total
quantity, and the total cost. Following are its transform and output:
out::rollup(in) =
begin
out.transaction_id :: in.transaction_id;
out.customer_name :: in.customer_name;
out.num_items :: count(1);
out.total_quantity :: sum(in.quantity);
out.total_cost :: sum(in.quantity * in.cost);
end;

178 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Reformatting to generic records
The Reformat to Generic component reformats each body record, adding a kind field with a
value of B and an ordering field with a value of 2. Following are the components transform and
output:
body::reformat(in) =
begin
body.kind :: 'B';
body.ordering :: 2;
body.* :: in.*;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 179

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The Reformat to Header and Trailer Generic component has two transforms: one each to
reformat the header and trailer records:
z transform0 Adds a kind field with a value of H and an ordering field with a value
of 1
z transform1 Adds a kind field with a value of T and an ordering field with a value
of 3
Following are the two transforms and their output:
transform0 (headers)
header::reformat(in) =
begin
header.kind :: 'H';
header.ordering :: 1;
header.* :: in.*;
end;

180 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

transform1 (trailers)
trailer::reformat(in) =
begin
trailer.kind :: 'T';
trailer.ordering :: 3;
trailer.* :: in.*;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 181

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

182 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Merging to header, body, trailer order
The MERGE component merges the records by using a two-part key {transaction_id;
ordering} to assemble the records in the right order:

AB INITIO CONFIDENTIAL AND PROPRIETARY Header and trailer record processing 183

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Reformatting to subrecord format
The Reformat to Header, Body, Trailer component reformats the data into the format defined in
ht1_input_subrec.dml.

184 Header and trailer record processing AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Geographic calculations
This chapter presents three graphs (provided in the distances sandbox) that show how to
calculate and use distances between geographic locations. Distance calculation is of obvious
importance in transportation and shipping, but it can also be useful in industries such as
banking for example, to find the bank branch nearest to a given customer's location.
The chapter starts with a graph that uses trigonometric functions in DML to calculate the
distance between two points on the globe (page 186). Then it presents two graphs that solve the
nearest branch problem mentioned above:
z The first graph does this in a simple, brute-force manner, calculating and comparing
the distance between every bank branch and every customer location (page 193).
z The second graph uses a more sophisticated and more efficient algorithm, which results
in much higher performance (page 197).

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 185

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Calculating distances between locations
The graph calculating-distances.mp uses DML to calculate the distance between two geographic
locations (also known as the great circle distance), given the latitude and longitude of each
location:

The graph also shows how to:

z Incorporate a package of reusable functions by using an include statement
z Use built-in trigonometric functions
z Use a fixed-point decimal type in the output record format to eliminate the need for
explicit formatting of numeric results

186 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

The INPUT FILE component contains the latitude and longitude of five pairs of cities in the
following format:

The record format is specified in shipping-routes.dml:

record
record
string(',') name; /* from_location name*/
decimal(',') latitude; /* Degrees latitude*/
decimal(',') longitude; /* Degrees longitude*/
end from_location;

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 187

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 record
string(',') name; /* to_location name*/
decimal(',') latitude; /* Degrees latitude*/
decimal('\n') longitude; /* Degrees longitude*/
end to_location;
end;

The OUTPUT FILE contains the names of each pair of cities and the distances between them. The
record format for this data is in route-lengths.dml:
record
string(',') route;
decimal(9.2) distance;
end;

Haversine formula and great circle distance

The REFORMAT component in calculating-distances.mp uses the Haversine formula to calculate
the great circle distance (in kilometers) between the two cities in each of the five pairs. The great
circle distance the shortest distance between two points on a sphere takes the curvature
of the earth into account. Since the earth is not a perfect sphere, the calculation is approximate
(but accurate to within 0.18%).
The REFORMAT calculates the distances as follows:
1. Incorporates the transform file great-circle-distance.xfr by using an include statement.
include "/~$AI_XFR/great-circle-distance.xfr";
The included transform file uses built-in trigonometric functions to calculate the distances:
constant real(8) radius = 6367.0;
/* Earth radius in kilometers */

constant real(8) to_radians = 0.0174532925;

/* Degrees per radian */

188 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 out::haversine(x) =
begin
out :: 0.5 * (1.0 - math_cos(x));
end;

out::great_circle_distance(a_lat, a_long, z_lat, z_long) =

begin
/* Convert inputs in degrees to radians: */
let real(8) a_lat_r = to_radians * a_lat;
let real(8) z_lat_r = to_radians * z_lat;
let real(8) a_long_r = to_radians * a_long;
let real(8) z_long_r = to_radians * z_long;

/* x should always be between 0 and 1, except for

floating-point roundoff problems that might knock it
out of that range: */

let real(8) x = haversine(a_lat_r - z_lat_r) +

math_cos(a_lat_r) * math_cos(z_lat_r) *
haversine(a_long_r - z_long_r);

/* Compute result: */
out :1: if (x <= 1.0 and x >= 0)
radius * 2 * math_asin(math_sqrt(x));
out :2: if (x > 1)
radius * 2 * math_asin(1.0);
out :3: 0;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 189

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 2. Outputs strings containing the names of each pair of cities, followed by the calculated
distance between them:
include "/~$AI_XFR/great-circle-distance.xfr";

out::reformat(in) =
begin
out.route :: string_concat(in.from_location.name, " to ",
in.to_location.name);
out.distance ::
great_circle_distance(in.from_location.latitude,
in.from_location.longitude,
in.to_location.latitude,
in.to_location.longitude);
end;

Simplifying the code and controlling the output by using the right data type
Because the graph uses the fixed-point decimal type decimal(9.2) for distance in the output
record format rather than a floating-point decimal type (such as decimal(9) or decimal('\n')), it
generates easily readable output data with no additional code to round the results of the
floating-point arithmetic.

190 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

If you run calculating-distances.mp, you should see the following output:

To see the effect of using a floating-point decimal type in the output record format for this
graph:
1. Save calculating-distances.mp under a different name (such as my-calculating-
distances.mp).
2. On the Ports tab of the Properties dialog of the Shipping Distances (in km) output file, select
Embed and click Yes when asked whether you would like to use the contents of the file
route_lengths.dml as the embedded value. (By doing this, you prevent this experiment from
overwriting the record format for the original graph.)
3. Change the record format of the distance field in the output file to the following and click OK:
decimal('\n') distance;

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 191

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 4. Run the graph and view the results. You should see the output below.
For delimited decimal types, the transform outputs all significant digits, in this case 12 or 13
after the decimal point, which is not very useful in this application; the distances would be
specified to the level of a tiny fraction of a millimeter. To avoid this, use a fixed-point decimal
type to explicitly specify an appropriate number of digits after the decimal point.

192 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Simple calculation of the nearest branch
The graph nearest-branch-simple.mp finds the nearest bank branch to each of a set of
customers by using a simple and straightforward approach that may be acceptably efficient if
the number of customers and branches is modest: it computes the distance to every branch for
every customer and selects the shortest one.

Here is a description of what nearest-branch-simple.mp does:

1. Uses INPUT FILE components that do the following:
a. Specify the record formats for the locations of the branches and customers:
branches.dml
record
decimal(",") branch_id;
decimal(",") latitude;
decimal(",") longitude;
string("\n") place;
end

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 193

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 customers.dml
record
decimal(",") customer_id;
decimal(",") latitude;
decimal(",") longitude;
string("\n") place;
end;
b. Provide URLs to the data files branches.dat and customers.dat.
Here is an excerpt from branches.dat:

2. Uses a JOIN to combine the two input datasets as follows:

a. Sets the key parameter on the JOIN to empty: { } to combine every branch with every
customer.

194 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. Calculates the great circle distance between every branch and every customer. The
complete JOIN transform looks like this:
include "~$AI_XFR/great-circle-distance.xfr";

out::join(in0, in1) =
begin
out.customer_id :: in1.customer_id;
out.customer_place :: in1.place;
out.branch_id :: in0.branch_id;
out.branch_place :: in0.place;
out.distance :: great_circle_distance(in0.latitude,
in0.longitude, in1.latitude, in1.longitude);
end;
3. Uses a SORT WITHIN GROUPS to produce, for each customer_id, a list of records sorted in
ascending order by the distance to each branch:
a. Sets the major-key parameter to customer_id and the order to Ascending:

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 195

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. Sets the minor-key parameter to distance and the order to Ascending:

4. Uses a DEDUP SORTED component to get the shortest distance for each customer_id:
a. Sets the key parameter to customer_id and the order to Ascending.
b. Sets the keep parameter to first to keep the record with the smallest value (shortest
distance).
5. Outputs the data, producing records in the format specified by nearest-branch.dml:
record
decimal(",") customer_id;
string(",") customer_place;
decimal(",") branch_id;
string(",") branch_place;
decimal("\n".2) distance;
end

196 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Here is an excerpt from the output:

Efficient calculation of the nearest branch

The following sections describe a more efficient (and more scalable) solution to finding the
nearest branch for every customer. The algorithm is more complex, and the graph, nearest-
branch-quadtree.mp, runs in two phases:
z Building a lookup file of map grids (below). The first phase builds a lookup file that
divides a map of branches into a sequence of grids, starting with a fine-grain grid and
building up to a very coarse grid. The lookup file is indexed by the size of a specific
square in each grid and the x and y positions of that square.

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 197

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 z Using the lookup file to find the nearest branch (page 203)

Building a lookup file of map grids

The following figure illustrates the results of calculating the first four grids for a number of
branch locations. The code used to create the grids is described in What the transform does
(page 199).

198 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

What the transform does
The graphs NORMALIZE component builds the lookup file as follows:
1. Includes utilities.xfr, a transform that does the following:
a. Includes the transform for calculating the great circle distance (page 188).

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 199

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. Defines constants for:
The location of the center of the map (a point in the center of the continental
United States)
The locations of the latitudes of zero distortion
Other needed values computed from the above
An integer representing a square size (in kilometers) that is larger than the surface
of the earth
c. Converts a point_deg_type to a point_km_type.
d. Uses the point_km_type value and a square size to get the southwest corner of the
square containing the point_km_type.
The complete utilities.xfr transform looks like this:
include "/~$AI_XFR/great-circle-distance.xfr";

constant real(8) phi_0 = 40 * to_radians;

constant real(8) lambda_0 = 90 * to_radians;

constant real(8) phi_1 = 20 * to_radians;

constant real(8) phi_2 = 60 * to_radians;

constant real(8) n = (math_cos(phi_1) - math_cos(phi_2))/

(phi_2 - phi_1);
constant real(8) G = math_cos(phi_1) / n + phi_1;
constant real(8) rho_0 = radius * (G - phi_0);
constant int largest_square_size = 65536;

200 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 out::map_coordinates(in) =
begin
let real(8) theta = n * (in.longitude * to_radians -
lambda_0);
let real(8) rho = radius * (G - in.latitude * to_radians);
out.x :: rho * math_sin(theta);
out.y :: rho_0 - rho * math_cos(theta);
end;

out::containing_square(point, size) = begin

out.x :: size * decimal_round_down(point.x / size, 0);
out.y :: size * decimal_round_down(point.y / size, 0);
end;
2. Includes point-types.dml, which defines two data types as follows:
point_km_type Represents a point on the map, with coordinates in kilometers
type point_km_type =
record
real(8) x;
real(8) y;
end;
point_deg_type Represents a point on earth, with coordinates in degrees
type point_deg_type =
record
real(8) latitude;
real(8) longitude;
end;
3. Calculates 17 grids with increasing square sizes for each input branch location, converted to a
point_km_type by map_coordinates. The length of a squares side (in kilometers) is a power
of 2 from 1 to 65,536 (20 to 216). A square of length 65,536 km is larger than the earth's
surface and so is guaranteed to contain every location.

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 201

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The complete NORMALIZE transform looks like this:
include "~$AI_XFR/utilities.xfr";
include "~$AI_DML/point-types.dml";

out::length(in) =
begin
out :: 17;
end;

out::normalize(in, index) =
begin
let point_km_type location_km = map_coordinates(in);
let integer(4) square_size = 1 << index;
/*The size of a square, in kilometers, is 2 to the power of
its index.*/

out.location_deg.* :: in.*;
out.location_km :: location_km;
out.square_size :: square_size;
out.branch_id :: in.branch_id;
out.square :: containing_square(location_km, square_size);
out.branch_place :: in.place;
end;

202 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 Following is an excerpt from the output (Branch Lookup File):

Using the lookup file to find the nearest branch

To find the nearest branch, the graph nearest-branch-quadtree.mp uses a REFORMAT in
combination with the Branch Lookup File (described in the previous section) to search
successively larger grid squares until the closest branch is found.

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 203

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 The REFORMAT transform does the following:
1. Includes utilities.xfr (described in What the transform does on page 199), which in turn
includes great-circle-distance.xfr and has a function for converting map coordinates from
point_deg_type to point_km_type:
include "~$AI_XFR/utilities.xfr";
2. Includes a DML file that defines the record type used in the lookup file:
include "~$AI_DML/branch-lookup-type.dml";
The record type is defined as follows (see point-types on page 201):
include "/~$AI_DML/point-types.dml";

type branch_lookup_type = record

point_km_type location_km;
point_deg_type location_deg;
integer(4) square_size;
point_km_type square;
decimal(",") branch_id;
string("\n") branch_place;
end;
3. Defines a variable for an imaginary branch that is much farther away than any real branch, for
use in initializing the loop that finds the closest branch:
let branch_lookup_type far_away_branch =
[record
location_km [record x 1000000 y 1000000]
location_deg [record latitude 1000 longitude 1000]
square_size 1
square [record x 10000000 y 1000000]
branch_id -1
branch_place "Far away"];

204 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

4. Defines a function, four_containing_squares(point, size), that takes a point_km_type, and a
size, which is the length of a square in the grid built by the NORMALIZE (see What the
transform does on page 199). The function returns the southwest corner of the square
containing that point, and the southwest corners of the three other squares closest to the
point. These are the four squares that overlap a circle of radius size centered on point. (See
the figure after the code below.)
out::four_containing_squares(point, size) =
begin
let real(8) x = size * decimal_round_down(point.x / size, 0);
let real(8) y = size * decimal_round_down(point.y / size, 0);
let real(8) s = (real(8))size;
let real(8) x_off = if ((point.x - x) < (s / 2)) -s else s;
let real(8) y_off = if ((point.y - y) < (s / 2)) -s else s;

out :: [vector [record x x y y],

[record x x + x_off y y],
[record x x y y + y_off],
[record x x + x_off y y + y_off]];
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 205

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

206 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

5. Defines several variables. Following are descriptions of the most important ones:

VARIABLE PURPOSE INITIAL VALUE DETAILS / CHANGE PER ITERATION

position_deg Location for which the Location provided The transform does not change the
nearest branch is in the input record value of this variable.
sought, in degrees

position_km Location for which the Location provided The transform does not change the
nearest branch is in the input value of this variable.
sought, in kilometers record, converted
to kilometers

size Length (in km) of a side 1 Initially, this is the size of the smallest
of a square in the square to search in. The optimal value
current grid depends on the typical distance
between branches.
At the end of each iteration, size is
doubled, and the next iteration searches
the next-larger grid size.

closest_branch Location of the branch far_away_ Initially, this is the location of the
that is currently closest, branch imaginary branch defined earlier (see
as a branch_lookup_ Step 3 above), which is farther away
type than any actual branch. Whenever the
transform finds a branch that is closer,
it assigns the newly found branch
location to this variable. When the loop
has iterated to search the next-larger
grid size (see size above) and has not
found a branch closer than the current
one (see distance_to_
closest_branch below), the loop
terminates, and the value of
closest_branch is output.

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 207

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 VARIABLE PURPOSE INITIAL VALUE DETAILS / CHANGE PER ITERATION

distance_to_ Distance to 1000000 Initially, this is set to a distance that is

closest_branch closest_branch farther away than any actual branch.
Whenever the transform finds a branch
that is closer, it assigns the distance to
the newly found branch to this variable.
When the loop has iterated to search the
next-larger grid size (see size above)
and has not found a distance shorter
than the current value of this variable,
the loop terminates, and the value of
closest_branch is output.

Here are the definitions in the transform:

out::reformat(in) =
begin
let point_deg_type position_deg = in;
let point_km_type position_km = map_coordinates(position_deg);
let integer(4) size = 1;
let branch_lookup_type closest_branch = far_away_branch;
let real(8) distance_to_closest_branch = 1000000;
let point_km_type[4] four_squares;
let integer(4) j;
let point_km_type square;
let integer(4) count;
let integer(4) k;
let branch_lookup_type branch;
let real(8) distance;
6. Loops to search an incrementally larger square in the Branch Lookup File as long as the
current square size is smaller than the largest possible square size and the distance to the
closest branch is greater than the current square size divided by 4. If the distance to the
closest branch is not greater than the current square size divided by 4, the larger grid cannot

208 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 possibly contain a closer branch; the closest one has been found, and the loop can be
terminated. (See the figure below.)
while (size <= largest_square_size and
distance_to_closest_branch > size / 4)

a. Inside the loop, the transform calculates a vector containing the coordinates of the
southwest corners of the square containing the point, and of the three squares closest
to it, each of length size:
begin
four_squares = four_containing_squares(position_km, size);

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 209

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 b. Still inside the loop, the transform looks up the square_size and the x, y coordinates of
the four containing squares in the Branch Lookup File to see whether it contains a
corresponding entry (or entries) for one or more branch locations. If it does, the
transform calculates the great circle distance between the input position and the
branch location(s). If more than one branch location is found for a given square size, it
keeps the closest one:
for (j, j < 4)
begin
square = four_squares[j];
count = lookup_count("Branch Lookup File", size,
square.x, square.y);
for (k, k < count)
begin
branch = lookup_next("Branch Lookup File");
distance = great_circle_distance(
position_deg.latitude,
position_deg.longitude,
branch.location_deg.latitude,
branch.location_deg.longitude);
if (distance < distance_to_closest_branch)
begin
// Found a new closest branch.
distance_to_closest_branch = distance;
closest_branch = branch;
end
end
end
c. As the last operation in the loop, the transform increases the square to the next-larger
size:
size = size * 2;
end

210 Geographic calculations AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

7. Outputs the customer and branch data:
out.customer_id :: in.customer_id;
out.branch_id :: closest_branch.branch_id;
out.branch_place :: closest_branch.branch_place;
out.customer_place :: in.place;
out.distance :: distance_to_closest_branch;
end;

AB INITIO CONFIDENTIAL AND PROPRIETARY Geographic calculations 211

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

CITI CATE NA : S/N: MPFTP-3134-695317288.579005
Cook>Book graphs listed by component
and feature
Graphs listed by component
Following is an alphabetical list of the most important components used in the Cook>Book
sandboxes, and the graphs that use them.

COMPONENT GRAPH PAGE

BROADCAST global_sort.mp 57

quantile3.mp 95

CONCATENATE multiplying-data-with-departitioners.mp 18

quantile3.mp 95

DEDUP SORTED nearest-branch-simple.mp 193

DENORMALIZE SORTED quantile3.mp 95

AB INITIO CONFIDENTIAL AND PROPRIETARY Cook>Book graphs listed by component and feature 213

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 COMPONENT GRAPH PAGE

FILTER BY EXPRESSION ht_validate1.mp 169

ht_validate2.mp 174

multiplying-data-by-3-way-self-join-with-FBE.mp 19

FIND SPLITTERS global_sort.mp 57

FUSE benchmarking-with-checking.mp 49

GATHER multiplying-data-with-departitioners.mp 18

INTERLEAVE multiplying-data-with-departitioners.mp 18

quantile3.mp (Get Counts subgraph) 97

JOIN difference.mp 65

nearest-branch-simple.mp 193

quantile1.mp 90

quantile2.mp 92

quantile3.mp 95

JOIN (Self-Join) multiplying-data-by-3-way-self-join-with-FBE.mp 19

LOOKUP FILE nearest-branch-quadtree.mp 197

MERGE global_sort_serial_simple.mp 62

ht_build.mp 176

NORMALIZE fill-gaps.mp 34

multiplying-data-by-fixed-number.mp 14

214 Cook>Book graphs listed by component and feature AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 COMPONENT GRAPH PAGE

PARTITION BY KEY AND SORT difference.mp 65

PARTITION BY RANGE global_sort.mp 57

PARTITION BY ROUND-ROBIN quantile3.mp (Get Counts subgraph) 97

READ MULTIPLE FILES read_and_write_multiple_files.mp 144

read_multiple_files_simple.mp 140

REFORMAT calculating-distances.mp 186

difference.mp 65

ht_process1.mp 160

ht_validate2.mp 174

make_difference_metadata.mp 84

name-parsing.mp 6

nearest-branch-quadtree.mp 197

REPLICATE benchmarking.mp 45

benchmarking-with-checking.mp 49

difference.mp 65

ht_validate1.mp 169

ht_validate2.mp 174

quantile3.mp 95

quantile3.mp (Get Counts subgraph) 97

AB INITIO CONFIDENTIAL AND PROPRIETARY Cook>Book graphs listed by component and feature 215

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 COMPONENT GRAPH PAGE

ROLLUP audit1.mp 111

audit2.mp 115

audit3.mp 116

audit4.mp 119

ht_build.mp 176

ht_process1.mp 160

ht_validate1.mp 169

ht_validate2.mp 174

quantile1.mp 90

quantile2.mp 92

quantile3.mp (Get Counts subgraph) 97

SAMPLE global_sort.mp 57

SCAN fill-gaps.mp 34

ht_process2.mp 165

making-unique-keys.mp 21

quantile2.mp 92

quantile3.mp (Get Counts subgraph) 97

216 Cook>Book graphs listed by component and feature AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 COMPONENT GRAPH PAGE

SORT global_sort.mp 57

global_sort_serial_out.mp 62

global_sort_serial_simple.mp 62

quantile1.mp 90

quantile2.mp 92

SORT WITHIN GROUPS making-unique-keys.mp 21

nearest-branch-simple.mp 193

TRASH benchmarking.mp 45

difference.mp 65

WRITE MULTIPLE FILES read_and_write_multiple_files.mp 144

AB INITIO CONFIDENTIAL AND PROPRIETARY Cook>Book graphs listed by component and feature 217

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Graphs listed by feature
Following is an alphabetical list of important Ab Initio software programming features used by
the graphs in the Cook>Book sandboxes, and the graphs that use them.

FEATURE GRAPH PAGE

Ad hoc multifiles ad_hoc_command.mp 137

ad_hoc_explicit.mp 134

ad_hoc_parameter.mp 137

ad_hoc_wildcard.mp 135

DML conditional fields ht_data.mp 153

DML include audit4.mp 119

calculating-distances.mp 186

ht_data.mp 155

nearest-branch-quadtree.mp 199

nearest-branch-simple.mp 193

read_and_write_multiple_files.mp 144

read_multiple_files_simple.mp 140

218 Cook>Book graphs listed by component and feature AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 FEATURE GRAPH PAGE

DML looping benchmarking-low-level.mp 54

ht_validate2.mp 174

make_difference_metadata.mp 85

nearest-branch-quadtree.mp 203

DML reinterpret_as function audit4.mp 126

DML subrecords ht_data.mp 155

DML trig functions calculating-distances.mp 188

nearest-branch-quadtree.mp 199

DML vectors ht_process3.mp 167

make_difference_metadata.mp 85

final_log_output function audit4.mp 121

in components

Graph parameters ad_hoc_parameter.mp 137

benchmarking.mp 45

generic_difference.mp 80

make_difference_metadata.mp 84

multiplying-data-with-parameter.mp 15

Local variables name-parsing.mp 6

Metaprogramming make_difference_metadata.mp 84

AB INITIO CONFIDENTIAL AND PROPRIETARY Cook>Book graphs listed by component and feature 219

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

 FEATURE GRAPH PAGE

Parameterized subgraphs multiplying-data-with-parameterized-subgraph.mp 16

Simple string manipulation name-parsing.mp 6

220 Cook>Book graphs listed by component and feature AB INITIO CONFIDENTIAL AND PROPRIETARY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Index

A auditing
graph, testing 129
ad hoc multifiles sample application to audit 111
about 132 simple, counting records 113
explicitly listing files 134 using log files 119
using a command to list files 137 automated generation of record formats and
using a parameter to list files 137 transforms 79
using wildcards to list files 135
ad hoc output multifiles 138
add_fields function in metaprogramming B
example 85
benchmarking
add_rules function in metaprogramming
elapsed time 53
example 87
graph 45
advanced topics
graph with checking 49
building flat data into header/body/trailer
low-level operations 54
subrecord format
other suggestions for 51
reformatting to generic records 179
startup time 52
reformatting to subrecord format 184
equality checks
fuzzy 78
safe 75

AB INITIO CONFIDENTIAL AND PROPRIETARY DO NOT COPY Index 221

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

D E
data multiplier graph, parameterized 15 elapsed time, measuring for performance testing 53
data multiplier subgraph, reusable 16 equality checks
date and datetime types, equality checks 75 fuzzy 78
deciling safe 75
See also quantiling
assigning deciles 102
fixing decile boundaries 92 F
graph, simple serial 90
filling gaps 33
decimals, equality checks 75
final_log_output function 121
departitioners, multiplying data with 18
finalize function, SCAN component
differences, finding 65
to fill gaps 36
differencing records
to fix decile boundaries 92
duplicate key values 74
to get starting record count per partition in parallel
finding differences 65
deciling graph 97
graph
flat input data, validating 169
creating simple 68
flat records 153
generic 80
fuzzy equality checks 78
variations on simple 74
distances between locations, calculating 186
DML G
metaprogramming 85
using to generate record formats 85 gaps in data, filling 33
using to generate transforms 87 generating large amounts of data
dml_field_info_vec type in metaprogramming about 13
example 85 controlling output size with FILTER BY EXPRESSION
dml_rule_vec in metaprogramming example 87 component 19
duplicate keys 21 multiplying data by fixed number 14
with CONCATENATE component 18
with departitioners 18
with GATHER component 18
with INTERLEAVE component 18
with multiplier parameter 15

222 Index AB INITIO CONFIDENTIAL AND PROPRIETARY DO NOT COPY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

generating large amounts of data (continued) integers, equality checks 75
with reusable, parameterized subgraph 16 is_valid function in benchmarking graph 48
generic differencing graph 80
geographic calculations 185
global sort J
producing serial output 62
JOIN component, empty key on 90
simple 57
globally unique ID 24
great circle distance 188 L
length function, NORMALIZE component 41
H length_of function in metaprogramming
example 85
hardcoded values, replacing with parameters 107
linked subgraphs 104
Haversine formula 188
listing files
header and trailer record processing
explicitly 134
building flat data into header/body/trailer
using a command 137
subrecord format
using a parameter 137
reformatting to generic records 179
using wildcards 135
reformatting to subrecord format 184
log files, reading back and rolling up values from 125
header and body subrecords, generating separate
low-level operations, benchmarking 54
summaries of 160
header information, distributing to body
records 165 M
merging records to header, body, trailer order 183
make_field function in metaprogramming
example 85
I make_rule function in metaprogramming
example 87
initialize function, SCAN component
map grids, building a lookup file of 198
to fix decile boundaries 92
measuring performance
to get starting record count per partition in parallel
elapsed time 53
deciling graph 97
graph 45
input parameters, adding and setting 54
graph with checking 49

AB INITIO CONFIDENTIAL AND PROPRIETARY DO NOT COPY Index 223

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

measuring performance (continued) P
low-level operations 54
other suggestions for 51 parallel quantiling, parameterized 103
startup time 52 parameterized data multiplier graph 15
metadata generation graph 84 parameterized parallel quantiling 103
metaprogramming, DML 85 parameters, replacing hardcoded values with 107
multiple files parsing names 5
processing 131 partitioning by range of key values 60
reading 140 partitions, sorting 60
reading and writing 144 performance, measuring 45
multiplying data elapsed time 53
about 13 graph 45
by fixed number 14 graph with checking 49
controlling output size with FILTER BY EXPRESSION low-level operations 54
component 19 other suggestions for 51
with CONCATENATE component 18 startup time 52
with departitioners 18 presorted input 61
with GATHER component 18
with INTERLEAVE component 18
with multiplier parameter 15 Q
with reusable, parameterized subgraph 16
quantiling
See also deciling
N adding and exporting parameters 106
simple serial deciling graph 90
names, parsing 5 subgraph
nearest branch creating and inserting into test graph 104
efficient calculation of 197 testing 109
simple calculation of 193
NORMALIZE component
length function 41 R
normalize function 41
normalizing vector data 167 re_get_match function in benchmarking graph 54
re_index function in benchmarking graph 48

224 Index AB INITIO CONFIDENTIAL AND PROPRIETARY DO NOT COPY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

reading multiple files SCAN component (continued)
and writing to a single serial file 140 scan function
and writing to multiple files 144 to fill gaps 36
reals, equality checks 75 to fix decile boundaries 92
record formats to get starting record count per partition in
automated generation of 79 parallel deciling graph 97
using DML to generate 85 temporary_type 36
record tracking 113 self-join with FILTER BY EXPRESSION component to
record_info function in metaprogramming control output size 19
example 85 serial deciling, simple 90
REPLICATE component serial output in global sort 62
when not to use 54 sets of files, processing 131
why not to use in some situations 104 skew in globally sorted data 59
reusable data multiplier subgraph 16 sorting partitions 60
splitters, calculating and broadcasting 59
startup time, measuring 52
S string_filter_out function in benchmarking
graph 48
safe equality checks 75
string_substring function in benchmarking
sample size 59
graph 54
sampling data 59
strings
SCAN component
equality checks 75
finalize function
manipulating 5
to fill gaps 36
parsing 5
to fix decile boundaries 92
subgraphs, linked 104
to get starting record count per partition in
summary files
parallel deciling graph 97
limitations of 116
initialize function
using 113
to fix decile boundaries 92
to get starting record count per partition in
parallel deciling graph 97 T
temporary_type, SCAN component 36
Text Tracking, turning on 45

AB INITIO CONFIDENTIAL AND PROPRIETARY DO NOT COPY Index 225

CITI CATE NA : S/N: MPFTP-3134-695317288.579005

Tracking Detail, viewing 54
transforms, automated generation of 79

U
unions, equality checks 75
unique keys
making 21
using a SCAN component to generate 30
user-defined types, creating 140

V
validating input data
flat 169
vectors 174
vectors
equality checks 75
normalizing data 167
to hold body data 158
validating vector input data 174
voids, equality checks 75

W
writing multiple files 144

226 Index AB INITIO CONFIDENTIAL AND PROPRIETARY DO NOT COPY

CITI CATE NA : S/N: MPFTP-3134-695317288.579005